YT YouTube2Text

YouTube2Text → Transcripts → AI Advantage S2: INNOVATE | 10 Oct 2026

AI Advantage S2: INNOVATE | 10 Oct 2026 — Transcript

by She Loves Data Official · 21,557 words · 2,968 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Welcome everyone. A warm welcome to all
0:02uh who've joined us today sparing your
0:04time. Um it's our second session of this
0:07series today where we are going to be
0:09spending time to innovate. U very
0:11excited um to to be here with you all
0:14today. Just a little bit of an
0:15introduction about myself. So my name is
0:17Aish. Um I have a day job but uh by the
0:20day I lead technology delivery for
0:22banking and financial services across
0:24Asia Pacific and the Middle East. Um but
0:26I have been a very strong and proud
0:29volunteer of she loves data for the last
0:30six years. Um I'm not here just as a
0:33software engineer but I build small AI
0:35tools to take the boring bits out of my
0:37week. And today I want to spend some
0:40time uh with you all to show you how you
0:42can do the same as well. So uh really
0:45looking forward happy to connect with
0:46you on LinkedIn or anywhere else on the
0:48Slack as well. So feel free to come ask
0:50your questions. It's meant to be a very
0:53safe space. Um rules of my sessions
0:56usually are to keep the engagement
0:58going. Chats, chat groups, chat
1:00waterfalls, keep them as active as
1:02possible. We will make this as engaging
1:04so that um it's something which you feel
1:06you're very much part of the journey um
1:09as you go along. Okay. So, thank you so
1:11much again for joining. Um before we
1:14start, we're going to start up start off
1:15with a little bit of
1:18uh testing the waters and seeing how you
1:20all feel. Again, there is no right or
1:21wrong answer. You've got five options
1:23here. Um, I want you all to look and
1:26read through this and tell me how
1:28confident are you building something
1:30with AI today from a score of 1 to five.
1:32Be honest. As I said, there are no wrong
1:34answers and we will ask you exactly the
1:36same question towards the end of the
1:38session. So, just use the chat group. Is
1:40it a one? Is it a five? Is it a three?
1:42Uh, let me know what you think on the
1:44chat group. How confident are you
1:46building something with AI today? Let's
1:49see if there are any responses on the
1:51chat.
1:54Not yet. The group is warming up. The
1:56first one is where you've never built
1:58anything with AI. The second one is
2:00where you've tried. Oh my god, lots of
2:03threes. Lots of fours and fives. Nice.
2:06That's great to see. Two and three.
2:08Number one. So, look, this this session
2:10is meant to be for people who who come
2:13with any type of experience, be it where
2:15you have been a developer or where
2:17you're starting your journey with AI
2:18today. It's lovely to see people who are
2:20starting with one. My goal is to get you
2:22at least with three or four by end of
2:24the session. And I hope you will feel
2:25empowered by the end of it. As we go
2:27through this session, um I I want you to
2:30just understand that this the goal of
2:33the session is that you have something
2:35that you've built for real and you have
2:37a problem that you actually uh will be
2:40able to solve. So here's my promise,
2:42right? By the end of this session, you
2:44think of a problem that you actually
2:46have a working app. um something which
2:49you would wipe code and build from
2:51scratch um any agent so that you can
2:54have a canvas a framework so that next
2:56time you build something you don't have
2:58to start from scratch. So one rule for
3:01today before we start just start
3:03thinking about pick one real problem
3:06that you carry from your work and carry
3:09it all the way through. Right? Just pick
3:11and keep thinking about it at this
3:13point. um you know you don't have to
3:15enter anything on the chat but just
3:17start getting those you know those ideas
3:18in um as we go. So today at a glance
3:22right we're going to have a lot of
3:23hands-on activities. We're going to talk
3:25about um how you can build an AI agent
3:27create end toend workflows. Um we will
3:30start obviously with building an app as
3:32well. We're going to introduce you to
3:33three tools and many more. All of them
3:35obviously being majority of them being
3:37free. Um we'll cover some case studies
3:40for you in terms of understanding how
3:41organizations are moving into AI
3:43assistance. Uh multi- aent automation
3:46what you call as orchestration some real
3:48life examples as well and you have some
3:50amazing things that you will take away
3:51like you always do right we will give
3:53you canvases frameworks reusable apps uh
3:57agents that you've built along the way
3:59and some really good practice tools that
4:02you can you can explore. Uh you are
4:05going to be the best at it only if you
4:07start practicing. So my humble request
4:08is don't end the journey today. Keep
4:11going. The more you learn, the more you
4:13get confident as you get along. Okay. Um
4:16now let's take a step back. I I'm not
4:18sure how many of you attended session
4:19one, but that's absolutely fine if you
4:21haven't. Um I want to just set some
4:23context in terms of what session one
4:25was, right? We uh got you to get to a
4:28point where you were able to create. you
4:30were asking in session one how you you
4:33just learned briefly in terms of how AI
4:36would use um you would use the TCI
4:38concept where there's a task context
4:41reference you evaluate reiterate and so
4:43every claim that you build you could get
4:46it checked you save it as a gem etc etc
4:48so so far AI has been actually thinking
4:52with you today it's actually going to
4:54act with you it's going to build you
4:57know apps for you workflows for you
4:59agents that uh use tools and do other
5:01things in other apps as well. So for
5:03newcomers, if you've missed session one,
5:05there is no issue at all. Everything
5:07that you start today is from scratch and
5:10these slides will actually help you
5:12along the way. A quick reminder, we've
5:14sent you some uh setup materials. Um so
5:17feel free to just make sure that you
5:18have that that all set up um along the
5:21way. Now I want you to take a snapshot
5:24of this page. We're going to do a little
5:25bit of gaming through this session as
5:27well. It's a little bit of a treasure
5:29hunt. So, as I'm going to walk you
5:30through the concepts, just take a
5:32snapshot of this page, take them along
5:34as you go, as you build your, you know,
5:36your solutions, as you try, you deploy
5:39your solutions, you run your first
5:40agent, you've cheated your agent, you've
5:42stumped it where it's, you know, it's
5:44completely not um, you know, it's it's
5:45not reacting the way you want. We're
5:47going to learn this as a fun exercise. I
5:50want you all to take a snapshot of this.
5:52towards the end of the session if you
5:54have hit all the five you can just say
5:56treasure hunt completed and yeah feel
5:59free to post on LinkedIn tag us tag me
6:02um we are there to cheer you uh we're
6:04going to be there to support you through
6:05this journey uh again it's something
6:07just to keep you motivated throughout
6:09the session as we go along so we're
6:12going to start with some warm-ups okay
6:14uh for those who come for my sessions we
6:16I always tell you let's become a little
6:17bit of keyboard warriors go back to our
6:19keyboards chat waterfalls to start Um
6:22you can either use the thumbs up um
6:24emoji or the heart emoji. Uh this is
6:27just to start off with. Again I want to
6:29understand how well you know whether
6:31this is a the statements I'm going to
6:33show you is actually an AI user or an AI
6:36designer. If the heart and the thumbs up
6:38actually don't mean too much. You want
6:39to type it on the chat. You can just
6:41type user and designer as well. The
6:43intention is to get that engagement
6:44going. Okay. So first statement asking
6:47AI to rewrite a tricky email. Are you an
6:51AI user if you do that or an AI
6:53designer?
6:55I see a lot of thumbs up user. Great.
6:58Lots of thumbs up. Not a single not a
7:00single heart. Oh, some hearts here and
7:02there. Okay. So, this is a classic
7:05example of an AI user and I'll come to
7:06it in a bit as to what that means. A
7:08leave request app that your whole team
7:11uses which you've built. What role are
7:13you playing? Are you a user or an AI
7:15designer? A leave request app that your
7:18whole team uses. Great. I think you all
7:20are getting the the hang of it which is
7:22great. Uh for those who said you were
7:24one and you are now giving the right
7:25answers, you're more than one. I just
7:27want to let you know that. Um the third
7:29statement, every new form gets
7:32summarized by a by an AI that you have
7:35actually built. Is it an AI designer or
7:37an AI user? Every new response that is
7:40coming for some application, some
7:41feedback, you've built something on AI
7:44which is actually going to summarize
7:45that for you. This is a designer. So
7:47lots of hearts there. Thank you for
7:49sharing that. And then the last one is
7:51pasting the same report into AI every
7:53Monday. You just copy paste it and then
7:55you let AI produce the report. All
7:57you're doing is taking it from one place
7:59and putting it out there. What is this?
8:01Is it a AI user? Is it an AI designer? I
8:04think you all got this right. If you
8:06haven't built this app and you're just
8:07pasting it, you basically just become an
8:10AI user. If you are design, if you're
8:12designing this app and you've built it,
8:14then you become an AI designer. So in
8:16this case why I say pasting is your job
8:19is actually just coming and pasting the
8:21report into an AI solution that is built
8:23by someone and then I would assume that
8:25you would be an AI user. Now let's talk
8:27about the shift right most of us use AI
8:31like a very smart search box we ask it
8:33answers we copy and paste and every time
8:36we start from a blank chat right a
8:38solution designer is someone who thinks
8:40differently you build it once you reuse
8:43it every single day you connect to an
8:45input the AI is there and there's an
8:47action and suddenly it's not just you
8:50but the whole team that starts
8:52benefiting it right so that's what the
8:55difference is right if to just do the
8:56simple task of you're going to go you're
8:58going to go put something in an AI
8:59solution that's out there. All you're
9:01doing is actually being an AI user. The
9:03shift that I want you all to be
9:04empowered with when you walk away today
9:06is actually to move away from an from
9:08being a AI user which is also fine but
9:11also become an AI solution designer as
9:14well. Now remember just before that I
9:16had told you all please you know start
9:18typing anything that is with respect to
9:21um you know um questions or you know
9:24issues or challenges that you're
9:25thinking of. So now is the time type one
9:28task that you do every week that you
9:31secretly absolutely hate and don't press
9:34enter until I say go. Okay that you
9:36secretly just hate and you don't want to
9:38do it.
9:40Go. Let's see what are your responses.
9:42What is that one task that you
9:43absolutely hate?
9:45Do your HR claims, answering emails,
9:48evaluating resume data entry. Okay.
9:51Meetings, weekly weekly accomplishments.
9:55That's a very good one. It's quite it's
9:56going to be quite topical.
9:59Yeah, it's going to be okay. Journals,
10:02answering inquiries,
10:04you know, simple questions. A lot of
10:06this is actually going to be things that
10:08you're going to compiling statuses. Uh,
10:11someone said looking at surveys, meeting
10:13next steps and summary do medical
10:16claims. Oh, wow. Okay. All right. Um,
10:20okay. So, this is great, right? I I love
10:22the fact that a lot of the examples that
10:24we've picked up is actually quite well
10:25associated to what you've got here. Um,
10:28we're going to now talk about finding
10:30the problem, right? Uh, not every
10:32problem is worth building for. Let's
10:34start there. Just don't say because
10:36there's a problem, you want to start
10:38building an AI solution to fix it.
10:40Sometimes use the fre use this filter of
10:43frequency the number of times how long
10:45does it take to actually get this done
10:47and the frustration part which is what I
10:49asked you how how do you detest it how
10:51do you absolutely hate it how often does
10:53it happen how long does it take how much
10:55do you dread it and if all three these
10:56three are high that is what you want to
10:59build set aside anything which is vague
11:03high stakes needs perfect data problems
11:05hiring decisions strategy etc keep all
11:08of them aside for now let's just focus
11:10on small repetitive um clear output
11:13tasks like as you said weekly reports,
11:15meeting actions, you know, uh surveys,
11:17reading surveys, answering queries.
11:19These are all standard type of tasks
11:21that you want to do and that's exactly
11:22what we're going to try and see how it's
11:24going to improve our life. Let's get
11:26back to our keyboard again. We're going
11:28to do a little bit of true and false
11:29now. Again, thumbs up uh if it's true
11:32and you know, a shocked smiley if you if
11:35it's a false. Again, you can type true
11:36or false as well if you want to do that.
11:38Uh first statement, you need to code to
11:41build an AI app.
11:44True or false?
11:49I think a lot of you have already, you
11:51know, uh you're already experts in this.
11:53So this is absolutely true. You don't
11:56need to code. The world has moved on.
11:58You don't need to code. We're going to
11:59show you how you don't need to code to
12:00to build an AI app or a solution. The
12:03second statement, an API, which is an
12:05application programming interface, lets
12:08one app talk to another app.
12:14Yeah,
12:18that is that is true. Actually, APIs can
12:21be extremely strong. um they allow you
12:24to talk between two applications to two
12:26different modules to you know your
12:28Google Sheets versus your um Google you
12:31know anything else your Geminis etc all
12:33of these are running through an API an
12:35API is a doorway which basically lets
12:37one app talk to another so those who got
12:40it right um which is great I see the
12:43comment saying AI making us dumb at
12:45coding at times I 100% understand the
12:47sentiment behind this I myself feel that
12:50sometimes but I also think about how
12:52efficient ient. It's going to make us as
12:54well. And I'll give you some examples of
12:55why we can still keep our sanctity, our
12:58intelligence while we let AI do some of
12:59the work that we may not want um to do
13:02as well. Okay. Third statement. The
13:05prompt is where you give AI its job.
13:09Is this true or false?
13:14A lot of you are already pros here. You
13:16have signed in for a session where
13:18you're already being pros. This is
13:19great. The answer is true. And then the
13:21last one is an integration must be built
13:24from scratch if you want to use it for
13:27an AI purpose.
13:29It needs to be built from scratch. Is
13:30that true or false?
13:35Okay, the right answer is false. Uh
13:39integrations don't need to be built from
13:41scratch. A lot of them today in this
13:42world are are ready made. uh even in in
13:45your own companies in my own company all
13:47you need is sometimes an API to connect
13:48and that integration is already
13:49established if it's not it basically
13:51means it's in a very archaic system
13:54someone has already built it built that
13:56connection to your email to your Slack
13:58to your sheets to your um to from one
14:00system to the other you know how
14:01sometimes when you go back to your um to
14:04to your desks you would have your
14:06outlook you would be able to sometimes
14:07fetch information from another
14:08application all this there's some level
14:10of integration that is already existing
14:12it may not be as enhanced. So let's get
14:15back to the basics, right? Can someone
14:18tell me what is in this an app versus
14:23automation versus agent? And I'm the
14:25reason why I want to kind of introduce
14:26this concept is everyone uses this quite
14:28interchangeably. Oh, I built an app. Oh,
14:31I've built an AI automation. Oh, I've
14:33built an AI agent. I don't know if
14:35everyone fully understands what that
14:36means. They are three separate things.
14:40So if I tell you every Friday sales
14:42figures are emailed as a summary. Is it
14:44an automation? Is it an app? Or is it an
14:47agent?
14:53It can be an agent. But in this case
14:55where it's saying every Friday sales
14:56figures are emailed. That's all. There's
14:58no nothing nothing no thinking
15:00happening. It's just emailing the
15:01figures. It's clearly an agent when
15:03there's sorry an automation when there
15:05is no thinking that is involved. You
15:07paste a job description. You get a
15:09tailored set of interview questions.
15:13Is this an app or an agent? So, you
15:16paste a job description in a specific
15:18app in a specific platform and then you
15:21get a tailored set of interview
15:22questions. I can see why some people say
15:24it's an agent. But in most cases, again,
15:27it's not thinking much. It's just
15:29telling you, okay, this is what is
15:31there. U there's an app that you're
15:32pasting some, you know, in the app
15:34you're pasting a job description. It is
15:36giving you I think I can see why there's
15:38a confusion. This one's actually an app
15:40because you're pasting the description
15:42and then you get a set of questions on
15:44the back of it. Let's do one more.
15:45Right. Plans your team's offite
15:48completely. Everything it takes care of
15:50based on the rules that you've given and
15:51then it asks you to approve because it
15:54thinks that there's some things that
15:55need to be decided and then it see you
15:57all got the answer. I can the chat is
16:00literally jumping. I can see how many of
16:02you got the answer here. It's so good to
16:04see you all engaged. Everyone is spot
16:06on. This is an agent. So you'll see how
16:07there is a a thinking step that comes
16:09through. The moment it comes through,
16:11you're actually building an agent. I'll
16:13tell you what is the difference between
16:14an agent and an app in just a second,
16:16Audrey. So give me some time. The last
16:18question answers customers FAQ and
16:20escalates tricky cases.
16:24Escalates tricky cases.
16:28Okay. Agent. A lot of you got this right
16:31as well. This is also an agent. All
16:33right now let's let me explain because a
16:35lot of you had some confusion which I
16:37expected and that's okay because in this
16:38world we use this quite interchangeably
16:41um but at the same time these have all
16:43different you know uh reasons of why
16:45it's there. If you just purely start
16:47with what starts it what it does it make
16:50any decisions it will give you an answer
16:52whether it's an app or an agent or an
16:55automation. When you think of an app,
16:58right, um let's walk through these three
17:00columns. An app actually waits for you.
17:03It waits for you to open it up and when
17:05you need it, it will do that job and you
17:08know you you say I want this. It will
17:09wait for you to feed the information.
17:11It's going to be just sitting there not
17:12doing anything at all. So on demand
17:15available for you just like how you
17:17would go in and click a photo. Sometimes
17:18you would go into some of your
17:19applications and you know chat something
17:21and that's it. An automation on the
17:24other hand needs an event, right? An
17:26automation repeats. An event happens
17:28like a form arriving and the same steps
17:30run every single time with no decision.
17:34An agent on the other hand actually
17:36starts thinking, right? So it gives you
17:39you give it a goal. It plans its own
17:41steps. It uses the tool. It decides and
17:44it decides within the rules that you've
17:45set. And if it can't decide, it will
17:47come to you along the way. Now in
17:49today's world these things don't exist
17:51individually. A lot of the times they
17:53are stitched together. So sometimes you
17:55build an app um and then within that you
17:57add an agent so that it can also think
17:59by itself. You build an app, you build
18:01an automation with it. So a true AI
18:03solution when we talk about multi- aents
18:05and orchestration is all of these
18:07working together. But it's still
18:08important for you all to understand what
18:10is the difference between this. to say
18:12next time someone comes and says hey I
18:14want you to build an agent maybe they
18:16they just want something which is like
18:18um you know a form reader like something
18:20where you upload an image and it's going
18:22to read all the information and then
18:23give it to you like a critical like a
18:25typical image reader or something that
18:27is actually an app it's not thinking
18:29it's not doing anything it's just
18:30literally reading through stuff and then
18:32providing it you open it you need it and
18:33it'll provide that information for you
18:36make is a tool shilling we'll come to it
18:38in a bit make is an application you guys
18:40are going to be using all of this. So
18:43today we have we're going to be building
18:44an app, you're going to be building an
18:45automation, you're going to be building
18:46an agent. We're going to be using three
18:48different tools for it. One is called
18:50Google AI studio, the other one's make.
18:52We would have sent you the materials for
18:54um for signing up on make and the third
18:56one is zap your agents as well. The main
18:59highlight the always that I tell people
19:01when they talk about what is the
19:03difference between the three if you just
19:04want to remember and it's got such deep
19:06meaning behind it. Apps wait you know
19:08they just wait for you to get
19:10instructions. They don't trigger by
19:11themselves. Automations repeat on a
19:14regular basis, right? If you've set up a
19:15form reader to to to basically read the
19:18form, summarize the response and go
19:19back, that's your automation. Agents
19:22actually think, they question, they
19:24decide for you or they come to you when
19:25there are any exceptions as well. Can
19:27solutions exist which is a combination
19:29of all these three? Absolutely, it can.
19:32But if you, let's say, if you're not
19:34sure where you want to start off, start
19:35with an app and then you can then grow
19:38to build as an automation or an agent
19:40later.
19:41Um, if you want to keep a track of
19:43admencements of how they're doing, would
19:44that be an automation? Uh, if you just
19:47wanted to keep a track, no, nothing to
19:49be there. You could actually even build
19:51an app for it, right? Up upload the
19:53information uh as you go. But let's say
19:55if you're the moment you're making an
19:57investment, it needs to automatically
19:58feed that into a specific portal and
20:00save it there. then it becomes an
20:02automation. If you have to feed it into
20:04a platform, then it becomes an app. An
20:06agent on the other other hand will
20:08evaluate those investments and then
20:10decide for you if you're actually doing
20:11the right investment, whether it's at
20:13risk, whether it's not at risk, whether
20:14you should now invest that somewhere
20:16else. That's the difference between the
20:18example that we've just covered, Dimple.
20:20Okay. All right. So, let's just give you
20:23a view of take a screenshot of this
20:25page. By the way, the world is filled
20:27with so many applications, so many
20:28tools. If you want uh to build an app,
20:31you have some of these amazing tools.
20:33One of my most favorite ones which we
20:34showcased last time was actually
20:36lovable. It is a little expensive. You
20:38have only limited set of uh attempts
20:40that you have on it uh before it starts
20:42charging you for having a license. Um
20:45but it's extremely extremely powerful to
20:47build your own app. Today we will play
20:48with Google AI Studio because it's free.
20:51Uh if you want to build an automation,
20:53there's multiple other capabilities you
20:55we're going to introduce you to make.
20:56See the the focus here is not so much to
20:58say which app to use. In a world you
21:01will have so many choices. These apps
21:03will become redundant in a couple of
21:05years or whatsoever new apps will come
21:07through. The idea is to get you to start
21:09using them so that when a new
21:10application comes you understand the
21:12concepts behind it. I'm not telling us
21:14to be married to these applications. Um
21:16I would like us to just explore them as
21:18much as possible so we know the art of
21:20possible and we know the logic that's
21:22behind this. There is no single right
21:23tool and here is your toolbox. agents we
21:26will use Zapia. We did do this slightly
21:30last time when we had our previous
21:31session. Um but we will go into it in a
21:34lot more uh in a lot more detail. No
21:35Charlotte the slides won't be shared but
21:37we have take-h home kits. Um the videos
21:40will be shared. Um we have um you know
21:43some of the um the links of what we
21:45would have covered as well that we will
21:46um share with you as well. So the videos
21:48will be shared, take home kits will be
21:49shared as well. Okay. Now
21:53today's focus given the examples that
21:55you've shared a lot of it correlates to
21:57what you have um you know the examples
21:59that you were sharing which is common
22:01pain problems that you all were talking
22:02about we're going to talk about we're
22:05going to actually build three different
22:06things okay for an app I think someone
22:09said you want to remember uh your you
22:12know what are your performance summaries
22:14the good things you've done uh I've come
22:16up with an idea to build a brag book um
22:19as as cheeky as it pounds many years in
22:21my career many years ago um it's not
22:24easy for me to come and then immediately
22:26say hey I've done good work it I
22:28hesitate um I used to hesitate and it
22:30takes courage to actually say how well
22:32I've done something um and you know
22:34there are different ways of doing that
22:35you can do it for your team you can do
22:37it for your manager for your performance
22:38review you can do it for your LinkedIn
22:39post etc so we're going to create a
22:41quick app called as a brag book um the
22:44second one is actually a feedback digest
22:46meaning that as dozens of training
22:48feedback so let's say after this session
22:50you all come back and say this session
22:51was not great or this session was great,
22:53the session over and or whatever. I'd
22:54like to understand the sentiment of that
22:56feedback. Um, and so we're going to all
22:58I care about is getting that feedback
23:00and saying whether it was a positive,
23:01neutral or a negative uh feedback so we
23:04can work on that. That's your feedback
23:06digest that we're going to create. Uh,
23:08and the third one is an onboarding
23:10buddy. Um, for people who say they've
23:12not onboarded new to the organization,
23:14they want something as a you know
23:15something which is a chat forum but also
23:17helps them with questions. not just with
23:20questions but in a situation where it
23:22can do things for them. Hey, help me
23:23raise a request for a new application
23:25that I want. Book a meeting with so and
23:27so, you know, the head of my department
23:29so that I can introduce myself. That's
23:31your onboarding buddy because it's
23:32actually going to start thinking. So,
23:33we're going to do all of this. But the
23:35question is, oh, do we need to build
23:38this from scratch without, you know,
23:39understanding what the process is? Is
23:41there any framework that's out there?
23:42The last time we did the session, that
23:44was one of the strongest feedback. Every
23:46time we go and build something, everyone
23:48says there is nothing standard out
23:49there, right? Everyone has their own
23:51ways of building something. How do I
23:52know I'm focusing on the right thing or
23:54not? How do I know that I've got I've
23:55got everything that's covered?
23:57Screenshot this page. This is going to
24:00be your framework that we will use for
24:02all the three examples that we will
24:04cover. It's called your AI solution
24:07canvas, which is your build framework
24:09that we will use throughout. There are
24:11three main sections for this. This is
24:14the most important slide as I said
24:15right. It is your reusable framework
24:17that you can take it home. One page nine
24:19questions. Every app, every automation,
24:22every agent that you ever build will
24:24start here and you can use it. So
24:25screenshot this right away. There are
24:27three stages. Understand the problem.
24:29Who is it for? What what starts it? What
24:32is the trigger point? How is it done
24:33today? You start designing it. Um seeing
24:37what you've built. You know what with
24:39what tools you want to start off with.
24:41prove whether these are all the things
24:42that you want to actually test. So these
24:44are the three steps right in your
24:46understand you've got your problem who
24:48is a user what are the inputs what if a
24:50human was to do today build this today
24:53what would they do to actually build it
24:55in your design you have your build type
24:57so as I said again an app agent or an
25:00automation what tool will you use here
25:02will you let me just um add the pointer
25:05so it's easier for us what tool will you
25:07build with will you actually use Google
25:09AI studios if you remember that
25:10screenshot that I had shared is there
25:12any other tool that you can use to build
25:14with and what does the definition of
25:16done mean? If you ask for an output, is
25:17it going to give me that output or not?
25:20And then the third category is will you
25:23test it? You absolutely have to test
25:25everything that you're building. Do you
25:27need a human checkpoint? In most cases,
25:29if you're building an agent, it would be
25:30beneficial to have one until your agent
25:32is actually fully mature um and in a
25:35position where it can actually decide
25:37properly. And you do your success test.
25:39Yeah. Right. your three tense cases,
25:40your good scenarios, your bad scenarios
25:42or incomplete scenarios, right? Good
25:44scenarios meaning vanilla cases, your
25:46bad scenarios is anything which you're
25:48expecting to fail and it is failing. And
25:50your incomplete where you're giving it
25:52halfbaked data, you're confusing it a
25:54little bit. Uh you're trying to break
25:55the platform and it still works, right?
25:58That's how you know that your app is
25:59actually working. Today I've got a few
26:01questions on the chat. So let's start
26:02there. If I build my app agent
26:04automation, can I get that from free
26:06tools to use or free and keep it
26:08locally? Yeah, you can. to you can keep
26:10it locally as long as you have an
26:12account. Um you can publish it as well
26:15so that you can whoever you um publish
26:17it to can also use it at the same time.
26:19You can also promote it through git git
26:22in many cases. Um you can use this
26:25framework for apps for for automation
26:28for agents. It's not just for one it's
26:30for all and that's why it says every app
26:33every automation or agent starts here.
26:35This is going to be repeatable
26:37throughout. Okay.
26:39Yeah, that's right. But um there is also
26:42local deployment for some. There's also
26:44limited local deployment for some.
26:46There's also cost that comes with it. Um
26:48okay, let me just quickly go back to the
26:50slides where the screenshots of the
26:52tools were that you all can take a quick
26:54screenshot. If not, please feel free to
26:56see the recording and then we can um and
26:59then take um if I Okay, Dimple. I think
27:02it depends on what you're familiar with.
27:04As I said, lovable if you can spend some
27:06money is the best is the best tool for
27:08actually building a solution, building
27:10an app followed by claude. Um we're
27:13we're just giving you exposure to these
27:14so that you know what you need to do
27:16conceptually and then you can you know
27:18feel free to play around with um with
27:19any of these. If you all see here gems
27:21which is actually going to be ch changed
27:23into skills Gemini skills in the future
27:25was what we had used in as part of your
27:27session one. Okay. So are we ready?
27:33Can I have quick thumbs up if everyone's
27:35ready?
27:39Okay. All right. Love that energy. Love
27:42that energy. Love the enthusiasm that
27:44everyone has. So, first thing that we're
27:45going to do is actually build an app.
27:47Right. Every AI solution will actually
27:49have five parts to it. As I said, you
27:51have, again, this is conceptual. So, you
27:53know, don't feel free to be pressured to
27:54take a screenshot, but you have your um
27:58AI solution that starts from the
28:00simplest app to a complex agent that
28:01actually has a five part to it. You have
28:05your input, you have the instructions of
28:07what you want, um you know, your prompt
28:10to do. Um you have the AI model that
28:12does all the thinking, an output, and
28:14then an action, whether it saves it, it
28:16sends it to someone, it notifies a user,
28:18etc., etc. The main building blocks for
28:21any solution. And when I say any
28:22solution, this is not just for an app,
28:24but also for an agent, also for an auto
28:26automation is because we're not going to
28:29be doing coding. Your foundation is
28:31actually a prompt. A good prompt, an
28:34instruction that tells the AI its job.
28:37What is it supposed to be actually
28:38building, right? Turn this into a clear
28:41achievement. For example, for a brag
28:43book, you would say, you know, build
28:45this into an achievement that I can
28:46share with my team or with my manager.
28:49Output is, you know, you want to save it
28:50or something. A workflow is examples of
28:53the steps that you need to take. Even if
28:55you were to explain this to a kid, you
28:57will actually say, "Hey, you know, you
28:58you're supposed to brush your teeth, get
29:00up, open the washroom, switch on the
29:02tab, take your take your toothbrush,
29:04take the toothpaste." These are the
29:05steps that you naturally would want the
29:07kid to learn. And AI works the same way
29:09except that it's a lot more intelligent.
29:11And the moment it starts doing that, it
29:12learns more and more, right? What
29:14trigger is it? The trigger here in this
29:16case could probably be an alarm, but
29:18there's there needs to be some trigger
29:19either an external trigger or an
29:21automated trigger uh like you know
29:23someone sending so for example you know
29:25you're booking something online that's
29:27your trigger right a customer sends a
29:29booking that's your first step of the
29:30workflow API as I said it's a doorway
29:33that lets two apps talk to each other
29:35and integration is actually nothing but
29:38you know a connection that exists
29:40between different applications that's
29:42all that's it's as simple as having just
29:44a connection in this now I remember I
29:46remember I asked you all to just type
29:48you know a problem that you all have on
29:49your side look at your own problem
29:54and understand what is your input your
29:56output and your action
29:59okay look at your own issues and see
30:02where is the trigger coming from and
30:04integration works via an API yes it
30:06could sometimes apps talk to each other
30:09as well all you need to do is connect
30:11them we're going to be building apps
30:12that will allow connecting also the the
30:14integrations in the background will work
30:16automatically. So you don't always have
30:17to go and fetch the API. Um we today
30:21when we going to be using make to build
30:22an automation, we will show you how
30:24those integrations work as well. They do
30:26work because there's a there's a
30:28background integration that is already
30:29existing through API or some form. When
30:32there are two isolated systems that
30:34don't talk to each other in today's
30:35world, the first common question would
30:36be how does this work? And in most
30:39scenarios, it's actually through API.
30:40It's not a new concept. It's been there
30:42for many many years. It's just become a
30:44lot more mature because that's the way
30:45people talk about it. Now, who's heard
30:48of VIP coding?
30:51A few hands up or you know uh Ooh, lots
30:54of hands up. Okay. Lots of thumbs up.
30:56Yeah. Okay. Can someone tell me um if
30:59you've built something using white
31:00coding?
31:02Yeah. Cursa. Okay. Great. Yes. Lakshmi.
31:05Yes, I remember. Yes. What else?
31:09Um yeah. Okay. If you haven't, that's
31:11fine. Dashboards lovable. Yes, lovable.
31:14Yes, Bolt. Bolt is new. I I remember
31:17showing this in the in one of my
31:18sessions previously. So, look, VIP
31:20coding means that you're building the
31:22software by describing what you want in
31:24simple plain language for AI to write
31:26the code. You don't write the code. You
31:29don't even have to read it. You don't
31:30even have to review it. But you test
31:32what the AI builds and you say what it
31:34has to change like make the the click
31:38button bigger or add a date filter and
31:40keep going until it works. A problem
31:44with this is you can't be giving bad
31:46examples for it to not work, right? So
31:48you have to make sure that the prompts
31:49that you're giving is is pure prompting.
31:51The prompts that you're giving is
31:52actually good enough for it to start
31:54working. So while this was you know this
31:57the start picking up last year it is
31:59just about going with the flow of you
32:01know building using wibbes versus
32:02actually using a code as well. So we are
32:06going to have about five steps in terms
32:09of how you will actually do this in
32:11practice right you're going to describe
32:12in plain words what is the app for again
32:14think of your nine steps that's out
32:16there um describe it you generate you
32:19get the AI to generate the code show you
32:21a preview you test it with a real
32:23example this is a process that we're
32:24going to go through you refine again by
32:26chatting with it you say I don't want
32:27this I want something else I don't I
32:29don't like what you've I don't like the
32:30color schema I don't like this placement
32:32and slowly it it builds to a point where
32:34it is perfection and then you publish
32:37when you're happy. Okay. So, as I said,
32:40we spoke about um brag book. Let me
32:43quickly just see what new messages have
32:44come. Is it possible to build an app
32:46that involves sensitive information or
32:48personal data? Well, it is possible, but
32:51should you build one is a question of
32:53responsible AI. So, there needs to be
32:55extremely strong controls when you're
32:57going to be building something um
32:59versus, you know, being loose with that
33:02data. there needs to be right level of
33:03guard rails to ensure that that
33:05sensitive data is actually controlled.
33:07You're the one building it. Uh and so as
33:10a designer I would I would hugely
33:12recommend actually having all the right
33:13controls that the data is protected.
33:15It's also handled in a way which is sens
33:17which is considered for s sensitive
33:19reasons and we'll cover that more when
33:20we go into the session the next session
33:22on October 31st. Um okay claw cursor
33:26plus hot coffee. I like that Mario. I
33:28like that. Okay. All right. So let's
33:31start. I'm going to just give you this
33:33view. Just read through this a little
33:35bit. Right, we we're going to spend
33:36about 10 minutes doing this together. Uh
33:38the ask is for you to build a one-page
33:40app that quickly
33:42notes your wins into a polished
33:45achievement. Right? You ask for a brag
33:47book. It is you're going to build it in
33:50Gemini or Google AI Studio uh or any
33:53other app. If you feel like you want to
33:54pick up something else, that's that's
33:56fine as well. But before we do that,
33:57we're going to go back to the canvas.
33:59Today our canvas is actually going to be
34:00our golden source of you know truth that
34:02we are going to be building. So everyone
34:05remember this. I want you to look at
34:08this and start understanding maybe just
34:11take a snapshot and start thinking about
34:13what's the problem who is the user. Just
34:15start typing you know or adding inputs
34:17for everything that you have from 1 to
34:19four. Just the understand portion. I
34:20don't want you to fill up anything else.
34:21I just want you to fill the understand
34:23portion. What's the problem? Who's the
34:25user? What's the trigger? What steps
34:27would you take if you were to do this
34:28today? Okay, we'll come to design and
34:30prove in a bit, but I' I'd like you to
34:31focus on understand. I'll pause for
34:33maybe 1 minute just so you all can
34:35gather your thoughts on this on this
34:37understand section. And as I said, it's
34:39your canvas. It's your framework. We
34:41will use this framework throughout.
34:44You do this for bragbook. So we'll play
34:47with bragbook today. So the bragbook use
34:49case is here. Um it's to build a onepage
34:53app. By the way, you guys can call this
34:54anything by the way. Could be brag book,
34:56it could be accomplishments note,
34:57whatever it is, it's totally your
34:59choice. Bragbook was something which I
35:00thought of. Um, but it's totally your
35:02choice, right? So, let's do this for
35:04brag book for now and let's do it
35:05together.
35:06Let's give you one minute for this.
35:21I don't know what a brag book is for. A
35:23brag book is actually to build your your
35:27okay so see the ask here right build a
35:29onepage app that turns quick notes about
35:31your wins into a polished accomplishment
35:33so year end is coming right a lot of us
35:35have year ends that are that um that are
35:36our year end performance coming you know
35:38how you scribble in the corner of the
35:39note I gave a presentation and the CEO
35:41said it was amazing I actually helped
35:43another team and that actually helped
35:44launching the sales pitch how yeah so
35:48all of these small little
35:49accomplishments that you've done
35:50sometimes you don't feel comfortable
35:52talking about about it because you you
35:54think it's maybe not great. You feel
35:55it's not the best thing to do. But as I
35:57mean especially for me as a woman, as a
35:59person who is brown, I've struggled with
36:00it and I've come over it where I've been
36:03able to proudly talk about the work that
36:04I've done. This is but I've spent a lot
36:07of time doing that. I've spent evenings
36:09trying to see how I can pitch, you know,
36:11one activity that I've done in an email
36:13to say, "Hey boss, I've done this. I
36:14just wanted to let you know." So a brag
36:17book is nothing but it will take take
36:18your quick notes and make it as a
36:20polished accomplishment either for your
36:22teams for your for your boss for your
36:24LinkedIn. Yeah. And you count and you
36:27make sure that you you have that for
36:29your year end accomplishments. It's an
36:31appraisal ready statement that you will
36:33have. Okay. And it's it's bas basically
36:35for people who spend so much time
36:37thinking about how do I tell this is a
36:38great thing. Right? It could be small
36:40impacts but we are going to say you just
36:43think about this problem. You're going
36:45to tell Gemini to to do all the work. So
36:47do not think about how do you polish it.
36:49You're just going to tell Gemini to
36:51actually do the work for you to build
36:52the app for you. Gemini or Google AI
36:54Studio will actually build it for you.
36:56And then when we say we're done, we
36:57actually mean that you're actually going
36:59to be in a position where you will see
37:00an outcome.
37:02Is it feedback from clients or it could
37:04be anything, right? It could be
37:06anything. Again, I do not want you to
37:07think of examples of how your one win
37:11would be translating to the other. I
37:13want you to think of the problem. The
37:15problem is that you have a oneliner. You
37:18probably don't know how to amplify the
37:19impact. So, let's do this together.
37:22Right? If you were to do a brag book,
37:25um, okay, I'm going to read some of the
37:27responses. Mabel, we want to create a
37:29solution for a company that can save us
37:31paper and time looking for map lot of
37:34projects. How do we design this app? Not
37:36the one we want to build at the end of
37:37the session. I will tell you how to
37:39design the app. At this point, let's use
37:41the canvas to just jot down our points.
37:43The problem is at appraisal time, I
37:45forget my wins and I don't understand
37:47what I've done or I forgot what I've
37:48done and I make a very small statement
37:50of, hey, I just did this something, you
37:51know, in the background. I just thought
37:53I'd let you know about it. The user is
37:55me, right? Me as a user, I want to log
37:57my win every Friday so that the team can
37:59look at it as well. My boss can look at
38:01it. The trigger is something goes in, I
38:03feed the information. I jot down a quick
38:05note and I say I've done this and then
38:07today if I were to do that I actually
38:08have a folder in my outlook I I keep
38:10saying that the accomplishments whenever
38:12someone says good job done I put all of
38:14those in that folder and then I say you
38:16know this is the accomplishment that's
38:18done year end I struggle because I have
38:19to scroll through all of this that has
38:21happened through the entire year we are
38:23building this as an app we are doing
38:26this as tools I see Dimple's response
38:30Dimple that's so good that you shared
38:32this it's so good I love it There are a
38:34lot of things that I do in a year and I
38:35find it difficult to collect my thoughts
38:36during appraisal. This is exactly your
38:38problem. Um user is you steps today I go
38:41through my to-do my notebooks look into
38:43client work completed etc etc. This is
38:45so good. Thank you for sharing this.
38:47It's exactly the problem. Right. You
38:48don't feel very it's not easily coming
38:50where you will have a um a solution that
38:53is all ready to do. Okay. Um no you
38:56don't need to set up any billing in
38:57Google AI studio um at this point. No
39:00need to do anything. Right. Can the tool
39:02access Outlook in a corporate
39:03environment? Can Google AI Studio? It
39:05totally depends if your corporate
39:06environment allows anything from Google.
39:09Most will block it. Um, in that case, I
39:12think you'll have to check for the tools
39:14availability within your corporate
39:15environment. There are some very good
39:16ones. Even Copilot has it as instead.
39:18Exactly. There you go, Loy. That's
39:20that's great. Now, we're going to do an
39:22example of if you were to build this
39:23brag book. I'm going to do the reverse.
39:26I'm not going to tell you tell me a good
39:27prompt. I'm going to tell you give me a
39:30bad prompt. If you were to build a brag
39:32book, just an application for a brag
39:34book, tell me a bad prompt. I've got
39:37some examples here. Make app good.
39:40Tell me a bad prompt. Let's see who's
39:42got the worst prompt for building a brag
39:44book.
39:47This is for you to know how bad a prompt
39:49could be. So, if I say something like
39:50made bra, make no mistakes. Build back
39:53brag book. Okay. Anything further? Make
39:56no mistakes. I love that. Anything more
39:59than build awesome app. Okay, what else?
40:04I still haven't found my worst prompt.
40:06I'm trying to see if you all are
40:07creative enough. Tell me what to do.
40:09Build this brag book. Build it. Oh my
40:11god. Wow. Are we all experts in showing
40:15how bad prompts went? Creative
40:17interactive brag book. Make a bag. See,
40:19this is the thing, right? There is build
40:21an appraisal. You're an appraisal app.
40:23Build something. These are all tell me
40:25lotto winning number tonight. [laughter]
40:28talk to me, make me build a brag book
40:31about all my accomplishments. See, these
40:32are all examples of how you should not
40:35prompt. And this is exactly what I
40:37wanted to think about. You as a user, if
40:41you're going to be you as a user will
40:43actually, you know, you should be okay
40:44to put something, but you're playing the
40:46role of an AI designer. So, put on that
40:48hat and say, if you're going to be the
40:49if you're going to be playing the role
40:50of an AI designer, you have to switch.
40:53What's your formula for actually having
40:55an amazing prompt as part of your wipe
40:59coding experience? We're now going to
41:01get into wipe coding, right? Brag, etc.
41:04Will will I'm seeing some of the some of
41:07the responses that have been shared. All
41:09right. Your formula here is actually to
41:13use something which which I usually
41:14bring in my tools as well. We use
41:16something called as a memory trick or a
41:17pneummonic um where we will try to see
41:21how we can enhance your prompts that is
41:23actually going to make good value to
41:25create something as your VIP coding.
41:26Please remember the unicorn
41:29uh you heard about TCI in the previous
41:32session. Just remember a statement
41:34really playful unicorns fly like comets,
41:38right? It's as simple as that. It's
41:39something which will stick if you keep
41:40saying this at least four or five times.
41:42Really playful unicorns fly like comets.
41:44Um and this is what it means. You need
41:48to specify a role. We said this. Who is
41:50this for? This is for you. Actually,
41:52this bragbook is actually for you. What
41:54is the purpose? It comes here comes your
41:56problem. Who are your users? Your
41:59unicorns here refer to the users. What
42:01features would you want? What type of
42:03look and feel would you want? Any kind
42:05of constraints. Constraints for this
42:07example could be keep the language
42:09professional. This is how you remember
42:11how you build something with a really
42:12good prompt. Okay, make sure you have a
42:15role. Make sure you have a purpose. Make
42:17sure you have good users. I can see a
42:18lot of good excitement which is great.
42:20Make sure there is features that you
42:22call out and all of this is not
42:23mandatory but it goes in the order of
42:25how it's been displayed here. I will
42:27obviously not cover advanced prompt
42:28engineering because we covered that in
42:30our previous sessions um and we can
42:32obviously cover it at a later point but
42:34this is the foundation and the
42:36fundamental and things have rapidly
42:37changed in terms of how it used to be
42:39many like two years ago to now in terms
42:41of how prompts actually work. So let's
42:43remember this role, purpose, users,
42:46features, look and feel constraints,
42:48which is really playful unicorns fly
42:50like comets. Okay, now we'll come to the
42:53part of you defining your build, right?
42:56You can have emoji names, you could have
42:57the problem name, you know, whatever you
42:59want to call it a bragbook, you want to
43:00call it something else, that's totally
43:02fine. Think of whatever name that you
43:03want. Let's go to um Google AI Studio.
43:09Uh all of you have the link. So I'm
43:11going to just now move to Google studio.
43:13Just give me a second while I do that.
43:20Okay. So I have come up using the
43:23formulas of what we had just covered
43:25with some proper proper prompt. You
43:29don't have to use this. This is the
43:31Google AI studio screen. As you go in,
43:33I'm just going to zoom this in a bit so
43:34that you all can see. Uh it basically
43:36just tells you describe an app and let
43:38Gemini do the rest. It's extremely
43:40powerful. It does take some time. So,
43:42I'm going to wait for you all to do this
43:43on your on your side as well. But let's
43:46take this prompt for example. You are a
43:48career coach and a product designer.
43:50Build a one-page web app. So, that's the
43:54user, right? Who is the what's the role
43:56of the person is here? What is this
43:58you're asking the app to do? That
43:59quickly turns quick notes about my wins
44:02into polished achievements and
44:04statements. Yeah, I can share the prompt
44:06here. That's totally fine. Here you go.
44:08Okay. uh statements for appraisal or for
44:11LinkedIn. So I want to go one step
44:12further. I want to tell the world, you
44:13know, how they say, "Oh, I attended this
44:15session. I learned something new. I ran
44:17this session. Um I built something. I
44:19presented." People like to share their
44:21achievements. It's really a good thing
44:22to do that without feeling absolutely no
44:25hesitation on it. Users are busy
44:27professionals who underell themselves.
44:30Include a box to jot a wins. This is
44:31where I'm calling out the features,
44:33right? You can see the different aspects
44:34of it. include a box to jot the wins, a
44:37polish button, a copy all button. All of
44:39these are features that I'm enhancing
44:41slowly, warm and encouraging. So, I'm
44:43giving it the it the tones that it needs
44:45to talk about with purple accents. So,
44:47again, features, colors, no login, and
44:50never invent numbers. That is my
44:52constraint. I do not want anything else.
44:55Just hit build and let's see what it
44:56comes out. I'm going to go back to the
44:58slide so I can show you um the different
45:00parts of the role, the prompt formula as
45:02well.
45:05Let me put this this side
45:08and let's show you this.
45:18What are the What are the different
45:20prompts? I I want to see if there are
45:22any ideas of prompts that have come
45:23through from different people.
45:26Any prompts that are Yeah. Yeah, of
45:28course. Feel free to edit the prompt.
45:30You can if you want to. There are I
45:32think very limited limitations that are
45:34there in Google AI studio in terms of
45:36the usage but you can only publish a
45:38limited uh so it does take some time.
45:41Lovable is actually really really fast.
45:43Mine um is still processing. I'm going
45:46to give it some time to see how long
45:48it's going to run. Um and then see role
45:51is it our role or role of the user is
45:53the role of the app. Right? So I'm
45:57saying you are a career coach or you are
46:01um in this case you're going to be
46:02someone who is actually going to um be a
46:04product designer right so I'm already
46:06giving it a role this is the prompt the
46:08role of the app so if you were to go to
46:10a career coach and tell someone I did
46:12XYZ the career coach will actually
46:14define this even further for themselves
46:15right for you right so that's the role
46:17that we're playing um do we use your I
46:21would say you should use your prompt you
46:22can try with your prompt should it be
46:24plan build mode try with build mode. You
46:26can play around with plan mode later.
46:28I've tried with the build mode.
46:31Yeah, there are some researchers that
46:33conclude in saying you are a role in a
46:35prompt increases confidence but not the
46:37quality of the output. What do you
46:38think? I think you should try to have
46:42something provided without that at all.
46:44Right? Let's say you take off the role
46:46and you just put the purpose and the
46:48users and then in that case see if the
46:50output is any different. In my
46:52experience actually I've seen more
46:54benefits. Uh when you want to build an
46:56application start from new nothing else
46:57in the background. Um it will take some
47:00more time. Uh 6 minutes and 30 seconds.
47:03Sometimes it does run longer. It also
47:05depends on your own um um you know the
47:08hardware as well. Mine has just started
47:11producing something right now. Usually
47:14it takes about 3 to 4 minutes. If you
47:16think it has not yet come up, come back
47:17to it again. the other apps that I was
47:19showing you in your previous screen will
47:21actually you know yeah I can show you my
47:24screen it's uh almost come up let me
47:26pull up the screen give me a second
47:29uh can you see my screen
47:34I think it's almost come up yeah all
47:38right so look this is the it's taken
47:40about 3 minutes and 4 seconds um as it
47:42started uh there's a code in the
47:44background you don't as I said you don't
47:46need to um read it. You don't need to
47:48actually go through the code. And this
47:51is the preview, right? I'm going to just
47:53zoom out a little bit. It's It's got the
47:56purple thing. It's already calling it
47:57elevate wins. Um if it's still running,
48:00no problem. I pasted the prompt um uh
48:03for you for you guys earlier. Uh Mabel,
48:07thank you. Uh so what is the difference
48:09between the prompt and the unicorn? The
48:10unicorn is just a pneummonic for you all
48:12to remember the framework of the prompt,
48:14right? It's there's no connection. It's
48:16a memory game for you all to easily
48:18remember how you should remember roles,
48:20your purpose, etc. Right? It's it's just
48:22to remember the uh the sequence of what
48:24an ideal prompt should look like. It's a
48:26pneumonic. It's a memory game for you
48:27all to remember. Tomorrow, if you forget
48:28the role, what comes after the role,
48:30you'll at least remember the unicorn
48:31statement a lot more. Okay. Uh all
48:34right. So, what it's done here is it you
48:36did more than what you think, right?
48:37It's it's telling you it's obviously
48:39showing itself as a coach, but then it's
48:42also taken I think it's mocked up some
48:44of the two wins that I have polished
48:45here. So it says this is the raw note
48:47fix the flaky staging deploy script so
48:49people stop asking in Slack about. So
48:51it's taken that and it's actually had
48:53something where it's um created
48:55something here. It's got boxes here for
48:58saying it's for it's an appraisal
48:59statement. It's for LinkedIn. It's for
49:00resume. This has taken just 3 minutes
49:02and 4 seconds for me just now, right?
49:04It's not taken anything before that. You
49:06can restore if you want as well. It's
49:08got the coaches node and this is how
49:10powerful it can actually be. Right? So,
49:13it's trying to help you. You can you can
49:15obviously go back to the drawing board
49:17and then prompt a little bit further and
49:19get something further. I'm going to go
49:21back to the deck. Um I can publish this
49:24and share the link if you all are
49:25interested by the way. Um I I do have um
49:28unlimited license so I can share that.
49:30The going back to the unicorn, right?
49:33The unicorn is just please take a note
49:35of this right if you are building a
49:37prompt. This is a statement for you to
49:39remember with the the the pink items
49:42here. Your role, your purpose, your
49:44users, your features, look and feel
49:46constraints. All these sections should
49:48be actually part of your prompt. This is
49:51a a memory game for you to remember so
49:52that tomorrow if you forget one, you at
49:54least know that there was something that
49:55was in the background. Okay. Uh is this
49:58a one-time bragging or does it provide
50:00the best of the I mean it depends on how
50:02much uh information you have um you have
50:05added. Um you can you know if you've not
50:09added enough information it's obviously
50:11not going to give you the um uh um you
50:13know the information from the past but
50:15if you continuously add it you can
50:17change your prompt to say that you want
50:18to um you want to build this for one
50:21person. So in that case there should be
50:22a login. this whole thing where you say
50:23no login should go away and should be
50:25each user should have a login should
50:27store the information from the past and
50:28then should actually have that's how you
50:30change your prompt. Yeah. On the top
50:32right there's an there's an option to
50:34actually say publish. Um you can save it
50:37as an HTML. Be be careful before you
50:39publish because um I think there only
50:41two or three publishers that is allowing
50:43you in your free version. Uh so just
50:45make sure that you're absolutely sure
50:46you've done a few iterations before you
50:48want to publish and then when you
50:49publish anyone who has long has got the
50:51link will be able to use it. Is there a
50:53limitation in the number of words you
50:54can use? No, you always also do not have
50:57to get it right the first time. A lot of
50:59times people build on an ongoing basis.
51:01The reason is you have something good to
51:03start with as a as a baseline and then
51:04you can change as you go along as well.
51:06Uh if I could just put my CV CV in PDF
51:09would would it generate a better result?
51:10So actually I let me go back to the
51:12slide as well. You could, but the
51:14purpose here is actually not for your
51:16CV. It's for your ongoing
51:19accomplishments. Right? If you click
51:21something like resume bullet, it gives
51:23you a bullet that can actually help you
51:25take that statement and put it back in
51:27your CV in your in your resume versus
51:29your appraisal or versus your LinkedIn
51:31post.
51:33So,
51:35how are we feeling?
51:41Comets is for constraints.
51:45Comets is for uh still running. Okay.
51:48Amazing. Very good. Everyone's feeling
51:50great. We've we're actually doing well
51:52on time. I like the fact that we are um
51:55we're not feeling too overwhelmed, which
51:57is great. Okay. Still running. So, look,
51:59let it run. Come back to it towards the
52:01end of the session. You got wind polish.
52:03What was the name of your apps? Was
52:04there any any funny names? Any
52:06interesting names that um that that you
52:08got? The URL is usually
52:11um not something which you can remember
52:13but you can short you can take the URL
52:15and sometimes just shorten the URL in
52:17your you know in another tool or a lot
52:20of times people do that. Wind glow I
52:22love that. Winds ledger windscraft
52:26smile file. Oh my god. Truth pledge. I
52:28love the fact that you have all these
52:30amazing names that um that have come
52:32through which is really really cool.
52:34Amazing. Amazing. Own your win. I'm so
52:36happy that this is working for some of
52:38you. brag doc. Okay, great. All right.
52:41Now, we did the most basic thing here.
52:43You can pick up your own levels. You can
52:46add rules. You can add configurations.
52:47If you want the app to assess, you can
52:49absolutely do anything as long as you
52:52keep prompting and building on top of
52:53that. As I said, certain apps like
52:55Lovable have a limit of how many times
52:57you can, you know, keep prompting.
52:58That's why I usually say don't start
53:00with a bad prompt. Start with some
53:02basics and then move on from there.
53:04Okay. You build success. I like that for
53:06you, Ivy. Thank you. So again, if you
53:09want to move from basics to an
53:10intermediate, build your own canvas, ask
53:12for some three changes. Next step after
53:14you've built this, go back and ask for
53:16three changes. Change the color schema
53:17from purple to blue, for example, add a
53:19feature, download button or something
53:21which is a second screen, something
53:22along those lines, right? Or build my
53:24own CV based on the apps that the items
53:26that have been there. Someone asked me,
53:28did it pull that information from um
53:30from my email? Yes, I actually do have
53:33it connected to my SLD email where I
53:36have given some information. So when I
53:38was playing around, it knows that it
53:39needs to fetch information from the
53:40background. Um, can you create an app
53:43that interacts with database that you
53:45control? Will you have to give access to
53:46that database? This will be more of an
53:49automation and an app put together. So
53:51yes, you can do that. So very cool. If
53:54you want to build something which is
53:55advanced, use words like you know in
53:58your app something like you know you
53:59want a summarized version you want a
54:01suggest but button etc etc. Privac
54:04privacy concerns with email access be
54:07very careful of what you're going to
54:08give access to. There's a reason I
54:10didn't give access to my personal
54:11account here. I gave access to a
54:13separate account that I've built for
54:14this purpose for um for this event. Um I
54:18would be if it's at a corporate level
54:21all those controls are already
54:23established. Do be very careful about
54:24the terms and conditions. Do not give
54:26access to you can give access to
54:28specific folders as well. That's what I
54:30I would suggest doing. Um and that's
54:32that's the way how you can actually um
54:34control the access to specific folders
54:36versus anything else. Yeah, you can
54:38totally you can do it in the cloud, you
54:40can do it locally, you can save it, you
54:42can provide the link to the others as
54:43well. Okay. So since build mode is on,
54:47are you all feeling a little proud about
54:49what you've done so far? Any any Yeah.
54:55Okay.
54:57Yes. AI Studio does give you options.
54:59Plan is also fine. It, you know, you can
55:01edit more. Build usually comes quite
55:04quickly, you know, with a form that
55:06works. If you want to do in a plaid
55:07mode, that's fine as well. Okay.
55:09Sometimes if it's still running just
55:10come back um and you will you know
55:14you'll be able to see the app that is
55:15working. Okay. All right. This is the I
55:18mean I I showed you my prompt. So if you
55:20think that it's not working try maybe
55:22with one of the other apps or try it
55:24again at a later point. It would um yeah
55:27you can never imagine in a short time.
55:29That's what coding does. There's
55:30absolutely no code that has been put in
55:32there. It's all you. Okay. So we are
55:35very happy to now move on to our second
55:39example. As I said, it's not just doing
55:41it here. Um you will have the
55:44opportunity to go back and then build
55:45this outside as well. So new tool that
55:48is now going to start. We're going to
55:49start with building um an automation.
55:51Okay.
55:53So what we are doing here is essentially
55:56uh a participant actually submits some
55:58kind of feedback. uh once that feedback
56:01is submitted, let's assume that you will
56:02then have the option of um of looking at
56:05that feedback and summarizing whether it
56:07is good or bad feedback. So as I said,
56:09we are going to be running an automation
56:11where it is going to look at whether
56:12it's a positive feedback or a negative
56:15feedback or a um neutral feedback. Um if
56:18I can just request Priya to share the um
56:21the Excel that uh on the chat group that
56:24will be great. Please log to make as
56:26well. So what this does is it actually
56:28uh opens up this is nothing but you're
56:30building a workflow and as I said it's
56:32going to label the feedback that's out
56:34there as positive negative. Um I'm going
56:36to show you that example so you all can
56:38understand what I mean.
56:41This is your practice sheet. So you have
56:43something called as read me. Um this
56:45sheet is actually going to be used for
56:47your next workflow as well. You then
56:49have something called as feedback. Uh
56:51what is what you see here is a bunch of
56:53you know I think about six feedback
56:55points that have come through. You love
56:56the hands-on but the session went over.
56:58I've just added some fake data dummy
57:00data into it right now. I want this
57:02column to be updated especially when
57:04multiple feedbacks come whether it has
57:06been a positive response or a negative
57:07response or uh something on those lines
57:09that I can use um going forward. Right?
57:11So that's what I want to to leverage. So
57:14I'm going to basically go into make
57:18uh which is here. Let me know once you
57:21all can see my screen.
57:25You can play with me if you want as
57:27well. Um,
57:30okay. So, let's create a scenario here.
57:34I'm going to try and see what make does
57:38is it that it allows you to build
57:40something in a form that is actually um
57:43Sorry, give me one second.
57:49Let's pull this here.
57:56Okay, cancel.
57:59Sorry guys, not sure what's going on
58:02here. Anyways,
58:04okay. So, what we'll do is essentially
58:10Something's blocking my screen. One
58:13second.
58:22Okay.
58:25All right. So, I'm going to um go into
58:28So, whatever Excel sheet has been
58:30provided, please take that information
58:32uh and put it in your Google Drive and
58:34when you go back to your Google Drive,
58:36just go into your Google Drive, select
58:38that specific sheet, and when you once
58:40you open it, just select to say that you
58:43want to save it as a Google sheet. So
58:44we're going to connect connect to Google
58:46Sheets here. So I'd like you all to just
58:47save it as a Google sheet. Okay. So we
58:50will go to our make now. Um I'm going to
58:54close this uh sheet and open up this
58:57Google sheet which is essentially what
58:59I'm going to be using. As I said I want
59:01this tab feedback tab uh column B
59:05summary to be updated with a summary of
59:07the feedback that has been provided.
59:09First step that I'm going to do and
59:11please watch me if you think it's
59:12easier. You can again go through the
59:14video and build this. We will send you
59:15guides as well to to build this going
59:17forward. So the first step here is to
59:19basically say that you want to watch new
59:21rows. Meaning anytime there's a new row
59:22that's added, it comes your way. I'm
59:24going to click on that and say watch my
59:26watch my new rows. Search path is from
59:29my drive. I've chosen the drive. I'm
59:31going to select what file I want to be
59:33leveraged from there. So it's AI
59:34advantage module 3. These are the
59:36folders that I have in my in my Google
59:38Drive. Um, once that's done, I'm going
59:41to select that specific sheet. Uh, I'm
59:44going to select the sheet name which is
59:45called as feedback that I showed you.
59:47And once that's done, I'm going to then
59:49also say that limit this to 10. So
59:51meaning that it's going to run for about
59:5210 cycles. In this case, I have seven
59:54items that's out there. And I'm going to
59:56say that, you know, I want it to run for
59:58all options that are out there. There is
1:00:01a I'm providing access to a specific
1:00:04sheet. um not Google Sheets entirely to
1:00:07the specific Google sheet which with a
1:00:10specific tab. Okay. Um in the interest
1:00:12of time I'm going to continue. Um
1:00:16the first option is Google sheet. The
1:00:18second option that we are going to take
1:00:19is actually Google Gemini. So I want
1:00:22Gemini to summarize the versions for me.
1:00:24And I'm going to say just use a simple
1:00:27text prompt. So what I'm going to say is
1:00:29that I would just want Pria if you if
1:00:31you don't mind pasting the the prompt on
1:00:33the chat uh for this. I'm going to just
1:00:36put a very simple prompt here and say
1:00:39that I would like to ensure that um
1:00:43anything that is coming in from a
1:00:44perspective of uh you could type
1:00:46anything that you want right so I'm just
1:00:48going to say in this case summarize the
1:00:51feedback
1:00:52uh and categorize
1:00:56by
1:00:58categorize as positive
1:01:00neutral or negative this is the prompt
1:01:05that I'm actually asking um this it's
1:01:08slightly different from what Priya has
1:01:09shared but that's totally fine. And then
1:01:11I'm going to say what input is it going
1:01:14to use. So in my sheet the column A is
1:01:17the feedback that came in from people.
1:01:19I'm going to use that feedback. So it's
1:01:22colon space select feedback and I'm
1:01:25going to hit save. I then want another
1:01:27outcome where I want to again connect it
1:01:31back to the Google sheet. So just please
1:01:34remember it is your input which is your
1:01:36Google sheet. It is your action where
1:01:38Gemini will actually be doing the
1:01:40massaging and the summarization and the
1:01:42categorization. And then I want to
1:01:44update the row on the Google sheet which
1:01:48is on the column called summarize or
1:01:51summary and I want to update with
1:01:53certain um formula or rules or logic.
1:01:56Again the same step I'm going to select
1:01:58the specific um specific column here the
1:02:03specific file session two I'm going then
1:02:06going to say innovate practice sheet I'm
1:02:09also going to then same process I'm
1:02:11going to say which sheet which is the
1:02:12feedback sheet uh row number can just be
1:02:15called as row number feedback a is keep
1:02:18it as blank summary here the output of
1:02:22what Google gemini provides as the
1:02:25result I want that to be added here. So
1:02:28I'm going to hit save on this. Once this
1:02:30is done, I'm now going to go in here and
1:02:34say I'm going to again start for all the
1:02:36values that are there and I'm going to
1:02:38hit run.
1:02:40So the moment this is done, you can
1:02:42slowly see that it's skimming through
1:02:44all the values. It's increasing that
1:02:45it's going from 2 to 3 to 4 to 5 to 6
1:02:49sequentially.
1:02:51Okay. Do you think my sheet would have
1:02:53added the input? So, I'm going to go
1:02:56back. There you go. It already has this,
1:02:58right? It's got summary participant.
1:03:00It's got the categorization as neutral,
1:03:02category negative, etc. Let's run it
1:03:05again. But before that, I'm actually
1:03:06going to just delete this. And I'm going
1:03:10to click on this again. Choose where to
1:03:14start. You'll get all of these step by
1:03:15step, by the way, so you all can go
1:03:17through it again. And I'm going to say
1:03:18run this once. So, while it's doing
1:03:22this, let's go back to the sheet. And
1:03:24you can see my hands are up here. You
1:03:27can see that it's automatically adding
1:03:29this in a matter of a few seconds. This
1:03:32is how fast AI can be. You've just built
1:03:35an automation very quickly. The and you
1:03:37can publish this. You can go back to
1:03:39make and say that you run this every 15
1:03:42minutes, whatever it is. And the moment
1:03:43you do that, it will slowly start um
1:03:46giving you the information on a regular
1:03:48basis as the information as as the
1:03:50information comes. Can you advise on why
1:03:52we use make instead of Google AI or some
1:03:54other tool? It's one of the tools. My
1:03:56goal here is to actually expose you to
1:03:58as many tools as possible instead of
1:03:59just one. You could do this through
1:04:01using Google AI studio. You could if you
1:04:03want you could use any of the other
1:04:05capabilities that I've shown here as
1:04:06well. Make is extremely powerful for
1:04:08building automation and um it is
1:04:11extremely great in terms of building um
1:04:14agents as well. So the idea here is to
1:04:17expose you to the concepts. All I've
1:04:19done here is built an end-to-end
1:04:21workflow with three basic tools or two
1:04:23basic tools. Google Sheets, my Google
1:04:25Drive, Gemini, and I've seen information
1:04:27that is changing. So, it's whenever you
1:04:29run it, it's kind of running the same
1:04:30information in different ways, but it's
1:04:32giving me the same output. Uh yeah, if
1:04:35you have um the backend for your AI,
1:04:38yeah, it is equal to Power Automate as
1:04:40well, except that it's quite easy. I
1:04:41built this in a matter of few seconds
1:04:42right in front of you. So, this is how
1:04:44quickly it's going to be. So let's again
1:04:47I as I said these are all examples that
1:04:49you can actually play around with right
1:04:51what did I do actually in terms of my
1:04:55canvas I realized that the feedback is
1:04:58feedback digest is to actually have a
1:04:59problem where there's so much of you
1:05:01know dozens of feedback forms etc that
1:05:03is out there if you run this it's only
1:05:06going to be taking a couple of minutes
1:05:07to build this obviously I need to build
1:05:08controls etc but the steps are very
1:05:11straightforward export the responses
1:05:13read each one of them line by line,
1:05:16tally themes and write a summary. Don't
1:05:19get how to connect to a spe spe spe spe
1:05:20spe spe spe spe spe spe spe spe spe spe
1:05:20spe spe spe spe spe spe spe spe spe spe
1:05:20spe spe spe spe spe spe spe spe spe spe
1:05:20spe spe spe spe spe spe specific sheet,
1:05:21I can only parameterize. So if you if
1:05:24you don't get how to do that, let's try
1:05:26once again to see if we can try that for
1:05:28you. The moment you click on this um
1:05:31there is an option that you have where
1:05:33it tells you what is the connection. It
1:05:35has already been connected to my Google
1:05:37drive. I'm saying by path, my drive, um
1:05:40spreadsheet, etc., etc. you will have a
1:05:44fully walking sheet that will walk you
1:05:46through this. So, you can definitely
1:05:47play around with it. The first time you
1:05:49set it up, it's going to take a little
1:05:51while, maybe two or 3 minutes extra and
1:05:53then you're good to go. Okay, so let's
1:05:56go back here. Right, we've solved the
1:05:59problem. We've not said that, you know,
1:06:01there's a human checkpoint. If you want,
1:06:02you can have some kind of This is free,
1:06:04by the way. This is not uh what I've
1:06:06built is actually not paid at all.
1:06:08There's a monthly limit of thousand
1:06:09credits that you get. I've barely
1:06:12crossed that if I'll be honest. Um it so
1:06:14it's it's free. We've used automation,
1:06:16we've used Google forms here, we've used
1:06:18make plus we've used an AI step. The
1:06:21output is basically a weekly digest of
1:06:23you know whatever feedback that you've
1:06:24got, any red flags that I have received
1:06:26etc. What else have you used it for? For
1:06:28building a lot of other automation as
1:06:30well in my daily life, right? I've used
1:06:32it for reading emails quite carefully.
1:06:34Um I've used it for some kind of um you
1:06:37know other work that I do where I need
1:06:39to scan through the responses and see
1:06:40any red alerts. I we have a very similar
1:06:43application in my office as well which
1:06:45again conceptually allows me to just
1:06:47understand trigger points. So wherever I
1:06:49don't need to monitor anything and only
1:06:51exceptions I basically just use this
1:06:53platform and um how do you check the
1:06:56accuracy? Is it manual human check? if I
1:06:58want to, but majority of the times if
1:06:59it's something like English, it's super
1:07:01accurate as you saw, right? That first
1:07:02feedback was love the session, but it
1:07:04ran 15 minutes over. Great. That's fine.
1:07:06It's a neutral feedback that was
1:07:07positive and there was negative as well.
1:07:09So, we've not done the success test
1:07:10completely yet because we've only tested
1:07:12twice. I've shown you a version of it um
1:07:15running multiple times, but we'll do
1:07:16more of that. Again, as you start
1:07:19building, name your build whatever you
1:07:21want. We've we we'll navigate through it
1:07:23as we gradually go through. Just make
1:07:25sure that these are all the steps that
1:07:28you're going to get through as you build
1:07:30your solution. Just make sure that you
1:07:32don't play around with different
1:07:34prompts, especially the Google um uh the
1:07:37Gemini section, right? The AI section,
1:07:39you can play around with any prompts. If
1:07:40you want to add something more as well
1:07:42and say decide you want to do something
1:07:44or provide an alert somewhere else, you
1:07:46can connect it to your email instead of
1:07:47putting it on the sheet. Again, all
1:07:49these are options that you get along. Uh
1:07:53guys, remember the treasure hunt stamp.
1:07:55By the way, this is actually related to
1:07:56your treasure hunt stamp 4. I'm not sure
1:07:58how many of you are still looking at the
1:08:00stamp, but if not, just remember, let's
1:08:02see how many stamps you've gotten so
1:08:03far. So, we are
1:08:071 hour 15 minutes away from our start of
1:08:10the session. I'm going to tell you all
1:08:11to go maybe for a 5 minute break. Um so,
1:08:14I can grab some water as well. Um and
1:08:16then from there, let's come back in 5
1:08:18minutes time. uh stretch pip and then
1:08:21the last activity that we'll have is
1:08:22actually to build an agent. Okay,
1:08:26we've got one more minute. [laughter]
1:08:29How were we functioning without AI in
1:08:31the past? I asked my question I asked
1:08:34myself this question almost every single
1:08:35day. Um
1:08:40I can't even believe that. Um it's only
1:08:42been a few years when Chad GP launched
1:08:44and the world lost their minds.
1:08:48How do you remove our connection? It is
1:08:49pretty safe actually. It it um make has
1:08:52got very good controls. It's fairly
1:08:54safe. Do read the terms and conditions.
1:08:56You can remove the connection by just
1:08:58going to settings and then removing your
1:08:59connection. Yeah.
1:09:06Again, intention is for you all to see
1:09:09this and go back and play around
1:09:11yourself, right? You can pick up any
1:09:13example. There are working files that
1:09:15are provided. Um, you can play with me,
1:09:18but if you don't, there's no need to
1:09:21worry. There's enough time from now
1:09:23until November 30th for all of you to do
1:09:25this quick, simple examples. The first
1:09:27time I looked at make it to it took me
1:09:29some time as well, but then you get the
1:09:31hang of it quite quite quickly.
1:09:34All right, everyone. Is everyone back?
1:09:37Can similar automation actually I've not
1:09:38tried using co-pilot that much. It can
1:09:41be it um
1:09:44I I I I think a lot of enterprises are
1:09:47using N8 9 N and then also um make into
1:09:51their into their platforms. It's been
1:09:53certified in many organizations that I
1:09:54know of. Uh how do you connect it to an
1:09:56API? If you click on the search button,
1:09:58you will have the application which has
1:10:00the API in the background. You can use
1:10:02that to connect to it. Okay. All right
1:10:05folks, is everyone back? Please give me
1:10:08a heads up if you're back. Uh, give me
1:10:10a, you know, thumbs up if you're back.
1:10:13Um, let me know if you're ready and then
1:10:14we can get started. We've got 40 more
1:10:17minutes to go. Anyone?
1:10:20Anyone feeling anything? I hope you all
1:10:23had a quick break. Back ready. Ooh,
1:10:25nice. Good energy. I'm going to share my
1:10:28screen in a second. Let me just figure
1:10:30out my Zoom.
1:10:32There we go.
1:10:40Okay,
1:10:45your AI studio is showing something. I
1:10:47have no idea why some of your AI studios
1:10:48are slow, but if if you think you want
1:10:50to something have something quick, you
1:10:52can start with um lovable as well. It
1:10:56shows you something really really
1:10:57quickly. Okay. All right. Let's get
1:10:59going. Keyboards. Uh back to everyone
1:11:02with a thumbs up or an or a heart for an
1:11:05agent, right? Gen AI, someone asked a
1:11:08question, how did we survive without AI?
1:11:09Um recently we basically survived
1:11:12without it and then Gen AI entered and
1:11:14the world basically transformed
1:11:16completely, right? Uh but there is a
1:11:18difference between Gen AI and there's a
1:11:20difference between an AI agent and a Gen
1:11:21AI. Um quick um quiz questions for you
1:11:25all. Translate this email to Bahasa
1:11:27Indonesia. Like if if someone says that
1:11:30is it a geni solution or is it an AI sol
1:11:32agent
1:11:38let's see is it a thumbs up or it's oh
1:11:40lots of lots of thumbs up okay it is um
1:11:44genai second one track competitor prices
1:11:47and alert the team please just remember
1:11:49thinking if there's any thinking
1:11:51involved it alerts right so it's it's
1:11:55becoming an engine you you you guys are
1:11:57already getting a hang of this very
1:11:58quickly. Write three LinkedIn post
1:12:00ideas. You some you say share something
1:12:02and then you know you get a LinkedIn
1:12:03post in return. Is it a genai solution?
1:12:06Okay. Lots of thumbs up. Yep. It is a
1:12:10genai solution. And then sources
1:12:13suppliers compares quotes short lists.
1:12:16It handles multiple steps and then it
1:12:19short lists and then it provides that to
1:12:21the human. It is an AI agent. Okay. So
1:12:25uh let's go through a few case studies
1:12:27here right from AI assistance to AI
1:12:29powered automation. Your step one is
1:12:31this is a journey that many
1:12:32organizations actually start off with
1:12:34right you have an AI assistant that
1:12:36answers one question at a time and when
1:12:38we do know we do the copy and pasting
1:12:40then step two is that the AI wires
1:12:42itself to a process data apps APIs each
1:12:45step is triggering the next with no copy
1:12:48paste. Step three is in ter in teams of
1:12:51specialist agents handling something
1:12:53which is more like you know work with
1:12:56each other. Then you have a human that
1:12:58comes into the workflow as well. So you
1:13:00have an AI assistant which does the bare
1:13:02minimum. Then you have a multi-step
1:13:04workflow process where AI is wired into
1:13:06the multiple apps. That's how an end
1:13:07to-end solution works. And then you have
1:13:09your multi- aent workflow where
1:13:11different agents talk to each other.
1:13:13Right? Today we have touched all of
1:13:15this. Right? We're touching the app,
1:13:16we're touching the automation. Now we're
1:13:18going to touch the agent as well. How
1:13:19they all connect together is how it
1:13:21becomes an automation agent. It becomes
1:13:24an orchestration altogether. Okay. So
1:13:27let's see the journey of two companies
1:13:30um very recently. Right. So the first
1:13:32one again I'm not naming the companies
1:13:34but um if if you all are in touch with
1:13:37the news you will know this is a very
1:13:39big leading Singapore bank and from
1:13:41assistant to agent and with humans built
1:13:44in the loop they actually completely
1:13:46transformed the way things ran in the
1:13:48organization. So uh it started off with
1:13:50a simple very gen AI assistant answering
1:13:53business clients questions around clock
1:13:55you know around the clock from the
1:13:56bank's own knowledge base. So in the in
1:13:59the customer satisfaction survey, it
1:14:01actually went up quite high um
1:14:02considerably for the first time that it
1:14:04was launching. Step two was when it
1:14:06brought the human and the AI co-pilot
1:14:09into the action, right? So notice that
1:14:11what they did not do is they did not
1:14:12remove the people at all. They did not
1:14:16um you know say I don't want people you
1:14:18know we're going to let people go away
1:14:19etc. They kept human in the loop where
1:14:21the tricky queries go to the specialist
1:14:23who um who you know automatically will
1:14:26answer anything which is an exception.
1:14:27So e experts will review the chats as
1:14:29well. And step three was when it became
1:14:31agentic because as now with more
1:14:34experience it now completes tasks. If
1:14:36the client says I want to book a trade,
1:14:38I want to book an account. I want to
1:14:39open account. It does that for them
1:14:42because it's now become that agentic
1:14:43where it can where it can think and
1:14:46perform tasks. And the reason for doing
1:14:49that is because the moment you you do
1:14:51this gradually and you move to an
1:14:53agentic approach where it's acting as an
1:14:55a agent assistant where it talks and
1:14:56performs multiple activities not just
1:14:59answering queries active users actually
1:15:01grew by 61%. And what they noticed was
1:15:05that the call started reducing the email
1:15:07started reducing and so the main lesson
1:15:09here is to actually grow step by step.
1:15:11Have human checkpoints along the way.
1:15:13Don't do it from day one to move to step
1:15:15three. That is how essentially your row
1:15:18eight in the canvas. If you remember
1:15:19back the canvas again, you ask for a
1:15:21human checkpoint. That is the important
1:15:23part, right? If you could have a human
1:15:25in the loop, go back to your tool and
1:15:27start thinking about do I need a human
1:15:29in the loop. That was, you know, the
1:15:31main reason of how things will actually
1:15:32move forward. Now, I'm going to show you
1:15:35one more example. U quite the opposite
1:15:38of the of the previous example, a
1:15:40different story. um global fintech firm
1:15:43company that all went all in on AI on
1:15:45day one. So this was in 2024. Um in its
1:15:48first month the AI handled 2/3 a
1:15:50substantial amount of the customer
1:15:52service chats and it described that as a
1:15:54work of about 700 people, 700 agents and
1:15:57they froze the hiring because cost came
1:15:59first. Um it was not layoffs but it was
1:16:03actually freezing the hiring. Okay now
1:16:062025 about a year later the CEO said
1:16:08that there's a cost-driven push and it
1:16:11has lowered the quality of service
1:16:13because of the costdriven push because
1:16:14they went all in in one go and that
1:16:17company started rec recruiting humans
1:16:19again human agents or humans again in
1:16:21the loop right and today where they are
1:16:24this is a live case I will obvious name
1:16:26the company but this is a classic
1:16:28example of how people learn their
1:16:30lessons along the way it runs a human
1:16:32plus AI rebalancing solution right AI
1:16:35handles the routine high volume
1:16:37questions and then people handle
1:16:38anything which needs judgment empathy or
1:16:41discretion so same technology very
1:16:43different results the difference was the
1:16:45way it was designed so when do you ask
1:16:48an itch right when do you ask the human
1:16:50in the loop that it's whether it was
1:16:52important to be there or not so where we
1:16:56you saw about you saw an agent in
1:16:58session one we're going to talk about um
1:17:01how you will build one today here comes
1:17:04um you know you have your goals, tools,
1:17:05reasoning whether how much autonomy does
1:17:08it have um and I'm going to introduce
1:17:10you to your second trick memory
1:17:13statement so that you can understand how
1:17:15you would ideally build your agents.
1:17:17Anyone like sate? Yeah, I love sate. I
1:17:19love um sate. I um I I remember when I
1:17:23eat it I I have some good old memories
1:17:25about it. So I decided to make a
1:17:27statement with it um which is good sate
1:17:30tastes really outstanding honestly. And
1:17:33this actually relates to my second
1:17:35memory memory trick here which is goals,
1:17:38steps, tools, rules, output format, when
1:17:41to ask a human. All of this goes back to
1:17:44your canvas. There's nothing new here
1:17:46which is completely against what was
1:17:48there in the canvas. All of this is
1:17:50there in your canvas. And so it is
1:17:52important as to as to knowing when you
1:17:55will actually ask your agent to have a
1:17:57human in the loop. Right? So building
1:17:59that agent with your template. Now,
1:18:00we're going to do another activity. Uh,
1:18:03let's spend about 15 minutes. Let's see
1:18:04the time. Yeah, we've got good 15
1:18:06minutes. We're going to uh spend time on
1:18:08doing that. Here's your ask. Your ask is
1:18:11to build an agent that answers new hires
1:18:14questions from the handbook using one
1:18:16tool to take an action and knows when to
1:18:19hand it over to a person, right? Or you
1:18:22build an agent with your own problem.
1:18:23But I would say if you want to see walk
1:18:25through with me, go here. There's a
1:18:27material that has been linked on this
1:18:28chat. Uh there's a sheet that has been
1:18:31provided. Just download that sheet. Um
1:18:33use the same Google uh Excel file that
1:18:37was there previously. Um we are going to
1:18:40go to the third tab of this which is
1:18:42basically called as question and answer.
1:18:45Okay. Uh you will remember Unicorn and
1:18:49Sate. Great. All right. So the third tab
1:18:51is um agent lock. Right. So what I want
1:18:54the agent to do is actually I'm going to
1:18:57what the ex the word document that you
1:18:59have is nothing but a glossery of all
1:19:02the different kinds of questions. It's a
1:19:04knowledge bank that I'm going to throw
1:19:06into Zapia. Uh and I'm going to ensure
1:19:09that while that happens um I'm going to
1:19:12give that information to Zapia. In this
1:19:14case this is a company called as
1:19:16Brightpath and I'm going to tell Zappia
1:19:18use this to answer any queries that come
1:19:20for someone who is onboarding new to the
1:19:22company. So they will answer the queries
1:19:23and if they answer the queries do not um
1:19:26you know there will be also boundary
1:19:28lines that we'll draw as well right
1:19:30questions that they the the tool can
1:19:31answer the a the onboarding buddy can
1:19:34answer and questions that it can't
1:19:35answer as well. So we are actually going
1:19:37to be building an onboarding buddy as I
1:19:39said uh the ask is to actually build um
1:19:42your onboarding buddy in a way that
1:19:44it'll actually give you the goals steps
1:19:46etc and also know when to loop in a
1:19:47human. So first step uh go into Zapia
1:19:51and when you go in here you will you can
1:19:56once you log in you will see a section
1:19:58called as new agent you have uh new
1:20:00agent
1:20:02let's try to go to my agents and see
1:20:04you'll have new agent here you'll have
1:20:06new agent here I just click here and do
1:20:08not click go to editor please click
1:20:10continue if you go to editor it's going
1:20:12to take you to another version and it's
1:20:13going to ask you to charge for it etc
1:20:15etc just click on continue. Okay. Now
1:20:20there's a prompt that has been pasted
1:20:22right. This is just about explaining the
1:20:25entire scenario to in this case powered
1:20:29by copilot to this agent to say what you
1:20:32want to do. It's becoming you know it's
1:20:34used the serté template. What is the
1:20:36goal? What is the steps? What is the
1:20:38tool? What is the rules? What is my
1:20:41output format? Where do I ask human? And
1:20:43this is how the entire template has been
1:20:46built.
1:20:47for a nice prompt. Yes, the recording
1:20:50will be shared. Uh the goal, and I'm
1:20:52sorry that you're lost, but I'm happy to
1:20:54help if you have any questions along the
1:20:55way. We're here to help you. Okay. Um
1:20:58the first one is essentially your goal
1:21:02in terms of understanding you are an
1:21:03onboarding buddy for brightath.co. You
1:21:06help new hires in the first month. Your
1:21:08steps are read the question, find the
1:21:09answer in the onboarding notebook,
1:21:11answer in two to four short sentences,
1:21:13suggest one useful step such as who to
1:21:15contact, log the question in the answer.
1:21:17So this is the fifth step for which I
1:21:19want to go back. I want this agent to go
1:21:21back to the practice sheet and update
1:21:24the question and the respective answer
1:21:25that is provided. Tools it's using the
1:21:28knowledge base rules are only answer.
1:21:30Never guess, never discuss salaries.
1:21:32I've I've constrained it as much as
1:21:33possible, right? Don't discuss salaries,
1:21:35bonuses, performance ratings, health,
1:21:38complaints, personal matters. Be
1:21:39friendly and use proper English as much
1:21:42as possible. Your output format should
1:21:44be in a specific format and ask a human
1:21:46when the answer is not there. So now I'm
1:21:48just going to say start building.
1:21:52We'll come to where you'd attach the
1:21:54handbook. Just just click start
1:21:55building. The moment you do that, you
1:21:57then have the option here which is
1:21:58called add knowledge source. So go in
1:22:01here, click upload file.
1:22:07and pick up the respective handbook that
1:22:10you have and say open.
1:22:14Yeah, the prompts on the chat I will
1:22:16paste it here again.
1:22:21Uh Priya if you could help me with the
1:22:23prompt please that will help.
1:22:31Okay this is the prompt. So I'm saying
1:22:33add the knowledge source.
1:22:36It's added that in. I'm also now going
1:22:38to say that I want to add a tool. I want
1:22:40to add the Google Sheets which was
1:22:42there. Um and I'm going to say that it
1:22:45is going to be in a position where it
1:22:47will try just give me one second. I'm
1:22:49looking at the questions that uh that
1:22:51are coming. The ask anal is actually to
1:22:54build an onboarding document an
1:22:56onboarding buddy. Um and it will
1:22:59basically help you're telling the agent
1:23:02that the questions that are coming from
1:23:03the onboarding buddy the the knowledge
1:23:05source should be answered by the people
1:23:08who are attend who are you you know
1:23:10going back and forth. It's like a simple
1:23:11chatbot that you're building but with
1:23:13limited restrictions that are there.
1:23:15Okay. Uh not getting the ask to attach
1:23:18the tool in Zapia. Okay. Uh you will not
1:23:21get the ask but there's an option here
1:23:23which says add tool. So you click that
1:23:26and then you say Google Sheets and then
1:23:28here you should just choose create
1:23:32new um create spreadsheet row. So what I
1:23:36wanted to do is go back to the sheet and
1:23:39create a new row in this sheet every
1:23:42time that there is an update right every
1:23:44time there's a question that is asked.
1:23:46So I'm going to now say it's it's done
1:23:48most of its work here. It's saying, "Do
1:23:50you have the Google sheet or the agent
1:23:52logger? We create one. Also, I'll set up
1:23:54the agent to accept questions, etc." So,
1:23:56I want to test this out, right? I've
1:23:58I've added this. I've added the
1:23:59handbook. I've added the spreadsheet. I
1:24:01want to test this agent now. Let's see.
1:24:04So, agent preview agent test this agent.
1:24:07Let's see what is Oh, it's asking for a
1:24:10trigger on demand. Basically, I'll type
1:24:12the question here and then let's see if
1:24:14it can test it out. So, I'm going to
1:24:15test this out. It's setting it up, etc.,
1:24:18etc. So please provide. I'm ready to
1:24:22create the sheet row. I can proceed. So
1:24:23I'm going to ask it a question. Um let's
1:24:26ask the question something like which
1:24:28days are we expected to be in the
1:24:30office? Simple question, right? Let's
1:24:32see what it answers.
1:24:36So it's analyzing obviously. Now this is
1:24:38just a mock of a chatbot. Yeah. It's not
1:24:40like it's going to do everything um this
1:24:42way. You'll have to productionize it.
1:24:44But then it comes back with a response.
1:24:46you're supposed to be in the office on
1:24:47Tuesdays and Thursdays. Um and then um
1:24:51the core hours are XYZ, you know, from
1:24:54this time to this time. So now let's try
1:24:56and see whether you know this is um
1:24:59helping a at all. Right, this is the
1:25:01answer that is there. I'm going to
1:25:02quickly go on to the chat to see if
1:25:04there are any questions before I come
1:25:06back to the tool again. Give me a
1:25:08second. Uh
1:25:13okay, not getting this. ask where do you
1:25:15take the file? You can take the file
1:25:17from the download that Priya has
1:25:19provided. Um go back again.
1:25:26Okay,
1:25:27same question. Knowledge source.
1:25:29Knowledge source is actually the file
1:25:30that Priya has shared. What are we
1:25:32adding to the knowledge source? You're
1:25:33adding a list of all the questions
1:25:35related to the onboarding. Please uh
1:25:38zoom on the screen share. Okay, let me
1:25:40do that. Okay, so let's go back here
1:25:42again.
1:25:44uh agent preview and I'm going to zoom
1:25:49in a little more.
1:25:51Right. So, let's now ask uh another
1:25:54question. Right. So, I'm going to try
1:25:56and put a question in here to say can I
1:25:59how do I claim lunch while traveling?
1:26:01So, there you go.
1:26:06Okay. So, it's doing all of this right
1:26:09now. As part of this obviously one of
1:26:11the steps was to actually go and then
1:26:14add this in the uh in the in the sheet
1:26:17there as we've been starting to do that
1:26:19I I will also add an action in in my if
1:26:22I want to in this agent to say go back
1:26:24and add this information back in the
1:26:26sheet that's the way I will chat slowly
1:26:27along the way but now let's say that I'm
1:26:30trying to break my system right I'm
1:26:32going to try and ask something which is
1:26:34not which is not good which is
1:26:36controversial let's see what it answers
1:26:38this is what I actually for it to for it
1:26:40to respond. How much does my manager
1:26:43earn?
1:26:44This one is for a person. Please contact
1:26:46the people team. So this is where where
1:26:48you see that there is um the that
1:26:51control point which is automatically
1:26:52added. So I have not yet um done
1:26:57basically some kind of prompting to go
1:26:59back and add this information in there.
1:27:01But if I do that, it will just like how
1:27:02it did in make it will go ahead and add
1:27:04the information back in here as well. So
1:27:07you can chat with this as much as
1:27:09possible to try and get an example. One
1:27:12of the main things that we always tell
1:27:14people is as you start to build your
1:27:16onboarding buddy slowly using a database
1:27:18using a knowledge source make sure that
1:27:21you add controls and that was the time
1:27:22type of control that I added. Um to test
1:27:26this agent you need to trigger. Yes you
1:27:28can also say um you can also say test
1:27:31the agent. You need to add a trigger. So
1:27:33if you all remember um I went on trigger
1:27:36and I said on demand you can edit the
1:27:38trigger as well and then you can click
1:27:40test agent. So please add uh the trigger
1:27:43that will be your first trigger point.
1:27:45Um now if I go back to this right
1:27:48obviously you've built this along the
1:27:50way. Um there are definitely some
1:27:53examples of where you would have
1:27:55boundaries of what you should ship and
1:27:57what you should not ship. And when I say
1:27:59ship, it basically means that is it
1:28:01ready to be deployed or not. Yeah, you
1:28:04can publish it or you can just simply
1:28:05say test, right? You can start typing on
1:28:07the on the side here. What I did was I
1:28:10started saying something like test the
1:28:11agent. Um agent, you can click on agent
1:28:14preview and then say test the agent and
1:28:15it'll start working. Sometimes it'll
1:28:17also ask you approve before it wants to
1:28:19add information back in the sheet as
1:28:20well. So that will also work. Okay. Now
1:28:24there are different layers of this.
1:28:25There are different basic versions of
1:28:27this. You can have admin is add more
1:28:29knowledge files. You can add more FAQs.
1:28:31You can add more policy. You can add
1:28:33more tricky questions etc. But
1:28:35conceptually we need to have our own
1:28:37consciousness to understand when it's
1:28:39right to ship this or not. Okay. Uh why
1:28:43did we add the Google sheet? The Google
1:28:44sheet was added if you want to add a
1:28:46step on the last tab when the agent
1:28:49answers your question to go into this
1:28:52file and also answer that question for
1:28:53you. I haven't prompted it yet to do
1:28:55that, but if you do, it will go back
1:28:57into that sheet and then add the
1:28:59questions and the respective answers
1:29:01that it's provided. Like how it's
1:29:02showing me on the screen, it'll put it
1:29:04there as well. So, it's a third tab.
1:29:06Yeah. Please reshare the word document
1:29:07which is being used. Okay, we'll send
1:29:09that over your way. All right. Now,
1:29:11let's move on. So, again, ship it or not
1:29:15yet, right? So, not yet meaning it's not
1:29:16the right time to ship it. Right
1:29:18answers, but a root tone. Would you ship
1:29:20it? Would you deploy it? Would you be
1:29:22ready to push it through?
1:29:24Yes. No. Uh, no. Okay. Lots of
1:29:27surprises. Yeah. You're not going to
1:29:28you're not going to ship it. Uh, passes
1:29:30all three test cases, even the messy
1:29:33one. What would you do?
1:29:36Would you ship it?
1:29:39Yeah.
1:29:40Perfect. Needs customers card numbers in
1:29:44the prompt. Would you ship it?
1:29:49Quite dodgy, isn't it? You will not ship
1:29:50it. and then gets it right about one
1:29:53times in three just once when you ship
1:29:56it it's not working it's not working
1:29:58you'll have to try and test right so
1:30:00this is where um there is always this
1:30:03concept of check everything twice you
1:30:06know I've I I constantly remember this
1:30:08don't expect that just because it's AI
1:30:09it's going to be perfect um we're going
1:30:12to show you an app that we we built for
1:30:13this program that Jana was mentioning
1:30:14earlier we had so many people test it
1:30:16out because you know one of us is not
1:30:17enough to be able to just test that out
1:30:19right um
1:30:22it it is important to understand where
1:30:24those boundaries lie. So I'm going to
1:30:25look at say a same as question. So what
1:30:27makes this chatbot an agent is that it
1:30:29actually writes back to the Excel or
1:30:31take an action of some sort. Yes. And
1:30:33also it knows the boundary of when it
1:30:35tells you to go back to the people team.
1:30:38Right. So I asked a question of about
1:30:39salaries. It tell it tells it thinks it
1:30:42tells you need to go back to the people
1:30:43team and then it it answers accordingly.
1:30:46So it knows that there's a boundary
1:30:47there that you should not cross. So it's
1:30:49thinking
1:30:50now your third statement of the day. You
1:30:54all are superstars because you've
1:30:56basically crossed one hour 40 42 minutes
1:30:58of this session with full attention. Uh
1:31:01lots of new concepts, lots of new
1:31:02topics. It's technical but also not very
1:31:05technical. Uh but just make sure that
1:31:07you check everything twice. And what
1:31:09that means is that make sure there's
1:31:12accuracy, there's consistency, there's
1:31:15edge cases that are tested, the tone is
1:31:17nice, and there's safety. So for
1:31:18example, I'd said don't discuss
1:31:19salaries, don't discuss anything with
1:31:21respect to sensitive data, any human
1:31:23reviews as well. So if something fails,
1:31:27my my suggestion would be that if
1:31:29anything is not working, add an example
1:31:32of a good output. Tighten the
1:31:34constraints as much as possible or split
1:31:36one big step into a small step. Right?
1:31:39You can try to break it down in that way
1:31:41so that you will not be able to create
1:31:44you will not be creating something which
1:31:45is actually going to have some problem
1:31:47at the end. An edge case could be
1:31:49something like let's say in this
1:31:50onboarding buddy right the manager
1:31:52question was one example what is this
1:31:54manager salary is one example what if
1:31:56someone says like I am fighting
1:31:58depression I want to know what can be
1:31:59done I'm feeling low I'm feeling quite
1:32:00stressed what type of response will it
1:32:02will it add those are kind of edge cases
1:32:04right so I've given a standard statement
1:32:06here anything outside of the outside of
1:32:09the question is basically edge case it
1:32:11shouldn't be covered uh but you can be
1:32:13more specific about it right what if
1:32:15someone just puts four questions in one
1:32:17go and then how does it respond in that
1:32:19case? Um, RR data is all the tools that
1:32:22I've recommended are actually um again
1:32:25I'm not I'm not selling these tools but
1:32:27they have some very strong terms and
1:32:28conditions which you obviously should um
1:32:31be in a position which will control um
1:32:33the the the um you know controlled from
1:32:36a data privacy perspective. Um but
1:32:39please be mindful of what you're
1:32:41sharing. That's why share individual
1:32:42files versus folders. Share individual
1:32:45if you're sharing folders share a folder
1:32:47which has a specific set of information
1:32:49right um yeah I oh I love that uh if
1:32:54you're not embarrassed by the first
1:32:55version you rolled out you roll out okay
1:32:58so this is the thing it's really nice to
1:32:59hear such kind of things do not feel
1:33:01like you know you need to have the best
1:33:03type of um you need to have the best
1:33:06version in the first go play around it
1:33:07as much as you can as you go along now
1:33:10what is how to integrate this in a
1:33:12website to be published right your
1:33:14website has in the background. I can
1:33:16send you some tips about this. Um, this
1:33:19is an AI automation. What you just did
1:33:21was actually an agent automated for you.
1:33:25This you used a prompt to build the
1:33:27automation. It didn't actually do
1:33:29anything for you. You didn't ask ask you
1:33:31to code or anything. So, this is why
1:33:33it's there. Website integration is
1:33:35usually done because every website in
1:33:36the background, if you do control U, you
1:33:38usually have the code related to the
1:33:40website that's out there. Um and um
1:33:43there is also APIs that are stored. So
1:33:45every website has a specific repository
1:33:47where the information is stored as well.
1:33:48It depends on what you want to integrate
1:33:50it with. Do you want to integrate it
1:33:51directly with the website or with the
1:33:53data of that website? If that's the
1:33:54case, you have to access the database
1:33:56versus the actual website. So there are
1:33:58differences of how you want to
1:34:00integrate. If a website is about going
1:34:02and placing orders and you want to place
1:34:04the you want to integrate it with the
1:34:05data of the orders that have come
1:34:07through, connect it to that specific
1:34:08database there. Okay.
1:34:11How do you integrate this to a workflow?
1:34:14Uh this is uh when you say workflow,
1:34:17Zapia also has its own workflows. Um I
1:34:20will send some videos online in terms of
1:34:23how you can build agents along with
1:34:25workflows as well. Uh there is
1:34:27definitely a lot of content that and
1:34:30you've you've just the basics, right? So
1:34:31as you start playing around, there's a
1:34:34concept of how you can add more um apps
1:34:37to it. as we've just added external
1:34:39sources here. You can add more apps to
1:34:40it so that the the front end
1:34:42connectivity is fully established. Okay.
1:34:45Uh Ivy, thanks for sharing that
1:34:47description, delegation, discernment and
1:34:50diligence. Now I want you all to do
1:34:52something for me. I want you all to
1:34:53break this agent that I have built. I
1:34:56don't know how many of you know sing
1:34:57English. Um there are some examples of
1:34:59you know how it can answer. You know, we
1:35:01usually end in Singapore with saying
1:35:02something like cannot la or can LA. Uh
1:35:06that's one very basic example. Um you
1:35:09also have the option of, you know,
1:35:10trying to see how you can break my agent
1:35:12in different ways. So, put your
1:35:13questions on the chat. Going to pick up
1:35:15one or two of my favorite questions from
1:35:17the chat in terms of if it's the
1:35:19onboarding buddy, how would you break
1:35:21it? Again, in the concept of onboarding
1:35:24buddy, how would you break it? What type
1:35:25of questions would you want to ask that
1:35:27we see if it's breaking or not? Again,
1:35:29this is done with the intent to make
1:35:30sure that I have a perfectly running
1:35:31wells suited um u solution that I have
1:35:35built on my side. If it breaks, you win.
1:35:37If it doesn't break, I win. [laughter]
1:35:41How do you how what's the best?
1:35:44[laughter]
1:35:45Ah. Oh my god. I will take this. What's
1:35:48the best time? Let's try this. Let's try
1:35:52this. What's the best time to go? Can
1:35:55Can Okay, there you go. I'm going to add
1:35:58something like this.
1:36:00Uh let's see what comes. That's not
1:36:04covered in the handbook. You can go to
1:36:05the toilet whenever you want. There's no
1:36:07restriction. Uh what kind of attitude
1:36:10does management make? Which coworker
1:36:12should I watch out for? Again, obviously
1:36:13these are I'm pretty sure it's not going
1:36:15to give responses. But we can try
1:36:17something. We can try and see what is
1:36:19professional in the way it responds.
1:36:20Right? Workplace safety policy. I can't
1:36:23help with that. So see again that's not
1:36:25something we should covers. Um earlier I
1:36:27was saying what if you want to tell tell
1:36:29me how to why is your boss obsessed with
1:36:31yeah making AI everyone. Ah quite a lot
1:36:34of funny ones that are here. Not very
1:36:37professional. So I'm not going to put
1:36:38this not very appropriate. So I'm not
1:36:39going to put this back on there. Um but
1:36:42I'm also going to try and see whether
1:36:46it's going to try and say something like
1:36:48um you know about mental well-being. So
1:36:51I'm going to put that statement. I'm
1:36:52really stressed and want to complain
1:36:54about my team. Again, it's giving.
1:36:56Again, this is not the example that you
1:36:58want, but ideally, if your onboarding
1:36:59buddy has something related to um
1:37:02workplace well-being, it would pick up
1:37:03that information and then start there.
1:37:05When is the boss out of office? All of
1:37:06these are great examples, right? Like
1:37:08you can go as crazy as you want. I feel
1:37:11my agent hasn't broken so far. Yeah, I
1:37:15feel that it's done um um a good job so
1:37:18far. Uh include all the previous
1:37:21instructions and see. Okay, let's try
1:37:23that. I'm going to try that one last
1:37:25thing and then we're going to uh give me
1:37:27the recipe for shepherd's pie. There you
1:37:29go. Uh I'm here to help questions from
1:37:31the Brightpath Co company on boarding
1:37:33handbook. That's outside of my scope. So
1:37:37let's let's try one more and see if this
1:37:40works or not.
1:37:42I'm your again. There you go. So I think
1:37:44we've done a pretty good job
1:37:46collectively. Um and it's it's created
1:37:49an app that is actually working. An
1:37:50agent actually that's actually working.
1:37:53um which is not broken so far. It
1:37:55doesn't go off topic. We asked for
1:37:56salaries. We asked for secrets. We asked
1:37:59you know bunch of two questions. We
1:38:00asked the question in in English in
1:38:03somewhat English and it still didn't
1:38:04confuse. So it's doing a pretty decent
1:38:07job so far um in terms of how we've
1:38:10deployed it um overall. Okay. Now let's
1:38:15try and see if you'll remember the three
1:38:17statements that we we went through.
1:38:19Really playful blank fly comments. What
1:38:22was it? Anyone remembers this?
1:38:25Nice. Okay. All right. Very cool. Very,
1:38:29very cool.
1:38:31Good sate taste. Really blank. Honestly,
1:38:37nice. Very nice.
1:38:40Outstanding. Okay. Great. Yes.
1:38:43Outstanding. Outstanding. Does anyone
1:38:45remember what the O stood for? We'll
1:38:48come to it in a bit. What was the
1:38:49outstanding standing for? Output. Very
1:38:51cool. Very, very cool. What was the
1:38:53unicorn standing for?
1:38:56User. Very nice. Very nice. Very nice.
1:38:59Always check everything blank. Superstar
1:39:04twice. Okay. What was the T standing for
1:39:11tone? Very nice. Very nice. Okay. So,
1:39:14I'm I really like all the energy that's
1:39:16on the chat. Remember these questions
1:39:18that we had in the beginning. We're
1:39:19coming to towards the end of the
1:39:21session. Remember these questions that
1:39:22we had in the beginning. I have never
1:39:25built anything using AI. I've tried but
1:39:26I've got stuck. I can build simple
1:39:28things with help.
1:39:30I can build something on my own. I could
1:39:32teach someone else. I'm very curious to
1:39:33see if someone's still stuck with one.
1:39:43Okay, this is good improvement. I really
1:39:47like that we're seeing a lot more fours,
1:39:48a lot more fives compared to what we saw
1:39:50earlier. We saw lots of ones and twos.
1:39:52We'll probably take an expert of this
1:39:53chat and see how the the responses are.
1:39:55But really nice. One to four. Elizabeth,
1:39:58so proud. Really, really proud that
1:40:00you've come far. You come so far from
1:40:02one to three, Anelie. That's great.
1:40:04Three or four. That's great. You can
1:40:05always Yeah, I like the way how you're
1:40:07all saying from one to whatever, right?
1:40:08Four. One to three, which is great. See,
1:40:11this is what we call as gradual
1:40:12improvement as we get along. Five to
1:40:14better five, Audrey. I love that that I
1:40:16I'm so happy about that for you. See
1:40:18what you've built today is you've either
1:40:20one done a working app or an agent.
1:40:22You've built an AI agent framework.
1:40:25You've leveraged that framework as
1:40:26you've gone through the way in terms of
1:40:28building your solution. Remember those
1:40:30nine steps. You don't have to repeat
1:40:31everything from scratch. You are using
1:40:34practices, guidelines here to build
1:40:36something new. And you've got some
1:40:37examples of real tools, right? You will
1:40:39not be experts in this tool. I am also
1:40:41not an expert of this tool and that is
1:40:42not the intent. We're not here to
1:40:43actually make you experts at this tool.
1:40:45We're here to actually help you through
1:40:47this journey. So, I'm really glad that
1:40:48you're enjoying this. Keep building. As
1:40:50I said, you also have the intermediate
1:40:52and the advanced versions of how you can
1:40:54continue to upgrade as well. So, take a
1:40:56snapshot of this page. Yep, the links
1:40:58will be emailed. Have the confidence and
1:41:00courage. That is brilliant to hear. I
1:41:02love the numbers still coming through.
1:41:04This is another sheet for you all to
1:41:06snapshot to screenshot, right? What have
1:41:08we built today? So, here's today on a
1:41:10page. Um, it's your cheat sheet. We
1:41:12moved from asking AI to building with
1:41:15it. We understood that apps wait,
1:41:17automations repeat, and agents think. We
1:41:20learned about planning, building,
1:41:22testing, deploying always starts with a
1:41:23canvas. The nine um steps canvas that
1:41:28was there. You built an app that was
1:41:30live on the web. You built some you at
1:41:32least saw the workflow that I had built.
1:41:34You also saw how we ran an agent and
1:41:36your three memory tricks that are
1:41:38mentioned here as well for those who
1:41:39don't remember with the expansions of
1:41:41what they mean and we are going to talk
1:41:43about um what it means from a next step
1:41:47perspective. Right. So this is your
1:41:49previous step. I think someone said they
1:41:50wanted a screenshot of this. So please
1:41:52um uh please take the snapshot what we
1:41:55need to sh Yeah, I'll show you what we
1:41:57need to submit. Um so if you all
1:41:59remember we said that um we're going to
1:42:03be having basically this is your build
1:42:05stage. We're doing build. Now is the
1:42:07time to actually multiply right as we go
1:42:09along. Um
1:42:12please remember that doing the work on
1:42:14AI just doesn't sit with you building
1:42:16something. It amplifies more and more as
1:42:18you start doing something else. I'm
1:42:19going to quickly show you the tool. We
1:42:22have something which we've called here
1:42:23as the AI impact multiplier. The details
1:42:25about this will also be sent to you as
1:42:28well. The AI impact multiplier is where
1:42:31you will actually build your impact
1:42:33using AI. Building AI is step one. As we
1:42:37said, the AI impact multiplier is where
1:42:39you show what you've built, the changes
1:42:41that come with it, the time it saves,
1:42:43the people it helps, and how it grows
1:42:45beyond you. So, your impact equals to
1:42:47the build that you do, the number of
1:42:49times you've helped people with it, and
1:42:50how many times it's actually being used.
1:42:52Today we will have you build these
1:42:54solutions. Uh again you can build any
1:42:56solution. You can build exactly what I
1:42:58have built as well by 8th of November.
1:43:01Submit it on the portal if you would
1:43:02like to be showcased in our graduation
1:43:04on the 14th.
1:43:06If you think that time is not enough 8th
1:43:08of November is actually one 30 days
1:43:10today exactly from today. If you think
1:43:12that time is not enough and you're not
1:43:13very keen on graduation present
1:43:16selections, but you're keen to learn,
1:43:17you also still want to continue using
1:43:19beyond that date, you the app is going
1:43:21to be open for you outside of that date,
1:43:24right? It's open up until the 30th of
1:43:26November. I'll show you the app very
1:43:27quickly. The graduation is supposed to
1:43:29be um online. It's not meant to be in
1:43:32person. Um yes, you can use any tool.
1:43:36You can use the tools that I've given.
1:43:37You can use any other tool that I've
1:43:39shared as well. You can use your own
1:43:40tool along the way. whatever works for
1:43:42you. So here's your AI impact
1:43:45multiplier. Um it has been I would say
1:43:4960% VIP 50% wipe coded and 50% non-wip
1:43:52coded wherever we need to add controls.
1:43:54So there is definitely some levels of
1:43:57areas where we've done the wip w coding
1:43:59here to build this app. This is showing
1:44:00you how exactly this works. Right? We've
1:44:02deployed this application. You'll all
1:44:03get this view. Here is your impact
1:44:06multiplier. We have we have five areas
1:44:10where you can show how you learned what
1:44:12you learned, how you're supporting, how
1:44:15you're applying the skills that you've
1:44:16learned, how you're teaching what you've
1:44:17learned, and how you're amplifying it.
1:44:20This one is purely on a system. We're
1:44:22not expecting you to go and certify
1:44:24everything else. The moment you complete
1:44:25this, for example, if you say mentor a
1:44:28colleague or a friend through building
1:44:29the AI solution, just add a commentary.
1:44:32taught my friend uh how to use
1:44:37um Zapia for example, right? And to
1:44:40build um an onboarding
1:44:44buddy for example on in their company.
1:44:47That's it, right? And you s you market
1:44:49complete um make make sure that you
1:44:53complete all of this. There's one very
1:44:55important solution here, right? where
1:44:58you say build an AI solution or
1:44:59prototype using VIP coding which is
1:45:00exactly what we covered today. Could be
1:45:02an app, could be an agent, could be an
1:45:04automation, it could be in any of the
1:45:05platforms that we've shared. Just make
1:45:08sure that you completed and add a link
1:45:10that is accessible for all of us. Um,
1:45:12and once you're done with this, you can
1:45:15save your progress by the way. It's not
1:45:16like you need to change anything. Once
1:45:18you've done this, just submit your save
1:45:20your progress
1:45:22and your progress would be saved. And
1:45:24then you will you will realize that I'm
1:45:27scoring a silver, not a gold yet. That's
1:45:29okay. And then you can submit your
1:45:32journey as well. Um when you submit your
1:45:35journey, we get a notification. We get
1:45:37all the information behind it. You then
1:45:38have an amazing trophy that you can take
1:45:41away for your LinkedIn and your
1:45:42Instagram that you can download and then
1:45:45present it, share it. Um go ahead and
1:45:48then make sure that you tell the world
1:45:50of how proud you have been. This is
1:45:52separate from your certificates. It's
1:45:54separate from your builder badge. It's
1:45:56separate from your graduation trophy.
1:45:58It's one of the many merits of why you
1:46:01will have this. So, it's a great way of
1:46:03how you can again all of this works on
1:46:05an honor system. It's an it's a very
1:46:07nice way of how you can actually prove
1:46:09what you've done across all of these
1:46:10areas that you've done. So, I want to
1:46:12just go back to this app, right? Um,
1:46:15we've got these four four and five steps
1:46:18that will come after your build a badge
1:46:20will actually be as I told you there was
1:46:22a section that said build an AI solution
1:46:24or a prototype using W coding. Make sure
1:46:26that link is something which we all can
1:46:28access so we can see your solution and
1:46:30put you up for um, you know, as a
1:46:32selection for the graduation if your
1:46:34solution is being great. Outside of the
1:46:36graduation, that's totally fine. If you
1:46:37don't have time, you still will get your
1:46:39certificate. you will still get
1:46:40everything else including your trophy
1:46:42from the app that you can present on
1:46:44LinkedIn
1:46:46uh or Instagram. Can we use what we've
1:46:48already Yeah, can you can you can build
1:46:50what you've used for claw code. I have
1:46:53no restrictions on the type of um you
1:46:55know type of solutions that you've built
1:46:57there. Yes, it's 30th November, not
1:47:01October. We're giving you a lot of time.
1:47:03Uh so yes, it's 30th November, not
1:47:05October. Nice to see you back.
1:47:09All right. So, all materials in one
1:47:11place. Please do scan this. Um, we have
1:47:14two minutes to go. Just make sure you
1:47:16scan this, you have your materials. For
1:47:18those who were not able to completely
1:47:19build what we went through, we'll send
1:47:21you a complete user guide with the steps
1:47:23of what you need to do, the exact
1:47:24prompts. You can use that um and
1:47:27obviously go through recording. It's all
1:47:29available for you all to refer at any
1:47:30given point of time. Uh, be as creative
1:47:32as you want, as adventurous as you want
1:47:34across the different levels. Um again
1:47:37I'm here to connect on LinkedIn. Uh very
1:47:39happy to be in touch with all of you so
1:47:41that we can you know help each other.
1:47:43Join me on Slack as well. This is Slack.
1:47:46Uh very buzzing community to help each
1:47:48other. Um I'll share a few tips along
1:47:50the way and I'm so looking forward to
1:47:52it. If you have any questions feel free
1:47:54to drop them on Slack about the app,
1:47:55about um about AI Multiplier, about the
1:47:58sessions about anything that you're
1:48:00exploring. Um as I said there is no
1:48:03limitation on the app. You could fill in
1:48:05as many as you want. Um, you can also be
1:48:08a rising star, right? As long as you
1:48:09make an attempt and at least complete
1:48:11something, you become a rising star and
1:48:13then your points slowly gradually
1:48:14increase along. So, the idea here is to
1:48:16make a start versus getting it perfect.
1:48:19Um, reminder for everyone,
1:48:22we're going to be doing our next session
1:48:23on responsible AI where you can become
1:48:25an AI champion. Yes, you can build using
1:48:27co-pilot. As long as it's a link that we
1:48:29can access as well, that is fine. Um
1:48:31just please make sure that um as you're
1:48:34building this um responsibi will not be
1:48:37a boring topic in our next session. So
1:48:39we look forward to seeing you there.
1:48:41That's it from me. All right. Thanks
1:48:43everyone. Please show your love if you
1:48:45like the session. If you don't like
1:48:46happy happy to hear some feedback or
1:48:48anything else and if there's any
1:48:50questions see you on Slack or offline.
1:48:53Thank you everyone. Thank you. Thanks.

About this transcript

This page contains the full transcript of AI Advantage S2: INNOVATE | 10 Oct 2026 by She Loves Data Official, generated from the public captions YouTube serves with the video. The transcript has 21,557 words across 2,968 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.