YT YouTube2Text

YouTube2Text → Transcripts → 4 AI Agents To Automate 99% Of Your Life

4 AI Agents To Automate 99% Of Your Life — Transcript

by Sandeep Swadia · 3,158 words · 480 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00four AI agents you can build today that
0:02will save you hours of work every week.
0:05You don't need any technical skills. If
0:07you can type, you can build them and
0:09start using them today. We'll do it
0:10together, step-by-step. For those who
0:13are new here, I spent more than 20 years
0:15as a CEO, board member, and investor in
0:18tech and AI companies, and I use these
0:21agents in my work every day. They're
0:23[music] game-changing. But, the agents
0:25are the easy part. You want to learn not
0:28just the tools, but the way of thinking
0:30or the system that changes how you work
0:32with AI agents forever. And that system
0:35works on any AI, Claude, Gemini,
0:37ChatGPT, you pick. We're going to build
0:39four agents around four pillars. I call
0:42it four C's framework: coordination,
0:45creativity, clarity, and coaching. Each
0:48of these pillars make your life less
0:50chaotic and more productive. So, let's
0:53get to the first pillar. The first C is
0:55coordination, because before anything
0:58else, AI should stop your day from
1:01getting hijacked. Microsoft studied how
1:03we actually work, and the numbers are
1:05brutal. The average worker gets 117
1:09emails a day, and we're interrupted
1:12every 2 minutes by an email, or a
1:15meeting, or a message. That's 275
1:19times a day. And by the way, none of
1:21that counts the social media
1:23interruptions. Our attention gets
1:25chopped into pieces every day. So, let's
1:29build a coordination agent that reads
1:31your inbox, checks the calendar, plans
1:34your day, and gives you the time and
1:36attention back. [music] We will start by
1:38building the email agent first, and then
1:41we'll add the calendar right after. For
1:44this walkthrough, I'm using Claude
1:46Cohaerence. Chat is where you ask
1:48questions. Cohaerence is where you
1:51assign work. But, don't get hung up on
1:53the interface. If you use ChatGPT or
1:56Gemini or any other agent tool, the same
1:59principles will work. The buttons will
2:01change, the workflow will not. So, okay.
2:04Here's the Claude app. You can select
2:06the mode,
2:07>> [music]
2:07>> whether you want to be in a chat tab or
2:10a co-work tab. Select co-work and it
2:13switches you into tasks.
2:15That's where we will build agents. Now,
2:19we want to think about email, so the
2:20first thing we're going to do is connect
2:22Gmail so the agent can reach and access
2:26your inbox. You do this from the plus
2:29menu.
2:30Connectors, add connectors, browse, then
2:34select. And congratulations, this is
2:36your first leap from chatbot to agent.
2:40Once Gmail is connected, you no longer
2:42have to copy emails into your chat. You
2:45can say, "Summarize my unread emails
2:47from yesterday." And the agent knows
2:49where to go and what to do. And by the
2:51way, if you're little uneasy about
2:54letting AI into your inbox, that's a
2:57good sign. Trust your hesitation. When
3:00you connect Gmail, Google shows you
3:03exactly what permissions you're giving
3:05it. Read it. Don't click through it
3:07blindly. For the first demo, I want the
3:10agent to read and draft emails, not send
3:14them. Now, we write the prompt. Same
3:17plain English you would type in chat.
3:19The difference is that now you're
3:21handing it a job. The prompt has five
3:24parts: the job, the tool, the
3:27categories, the output, and the
3:29boundary. You could say something like,
3:31"Review my unread Gmail from the last 24
3:34hours. Sort it into three buckets:
3:36urgent, informational, and ignore. For
3:39anything urgent, draft a reply that
3:42sounds like me. Don't send anything
3:43without my approval." Notice the five
3:46parts in that one sentence? The job, the
3:49tool, [music]
3:50the buckets, the output, and the
3:53boundary. Now, underneath every agent
3:56runs the same loop. It reasons, it acts,
3:59it looks at the output, and reasons
4:01again.
4:02Researchers have a name for it, react,
4:05reason and act. But, email is only half
4:10of our coordination agent, correct? Once
4:12we've used the email agent for a few
4:15days or maybe even a few weeks, we're
4:17getting comfortable now. We're ready to
4:20add the second coordination layer,
4:22calendar. Now, for the walk-through,
4:24same path as Gmail, plus menu,
4:27connectors, add connectors. This time,
4:30select Google Calendar to connect to
4:33Google Calendar. Again, I'm not giving
4:36the agent permission to move meetings or
4:39change my day without my approval.
4:42At least in the beginning. Here's what I
4:44would prompt. Look at my Google Calendar
4:47for today and tomorrow.
4:49Compare it against all the urgent emails
4:51that you found. Organize my day by
4:54telling me what creates conflict in my
4:56schedule, what needs an action or prep
4:59before a meeting, and what can wait.
5:01This is the moment the email and
5:04calendar start collaborating with each
5:06other. Now, the agent starts helping me
5:09shape my day. Once I run both these
5:12email and calendar versions a few times,
5:15>> [music]
5:15>> and I trust both outputs, I can schedule
5:18the full coordination run every morning.
5:21And from that point onward, every
5:23morning, you'll know exactly what your
5:25day is going to look like. In Coda, you
5:28do this by going to the schedule tab in
5:31the sidebar or using the schedule
5:33command inside a task. Of course,
5:35[music] be careful about scheduling
5:37anything that touches sensitive data or
5:40acts on your behalf without your
5:42approval. And then, as you get more
5:44comfortable, you can promote this agent
5:47as your executive assistant. You can
5:49prompt, "Can you block two slots for 30
5:53minutes each so I can go running with my
5:54buddy?" The agent will find time for you
5:57so you can do things you want to do on
5:59your busy day. So that's your
6:01coordination agent. The first lesson,
6:05don't delegate the decision immediately.
6:08First, make the work visible, then make
6:11it efficient, then make it automatic,
6:14and then and only then use your judgment
6:17to delegate and hand off slowly. [music]
6:20And the second lesson is to not hurry
6:23through this process.
6:25Start with email,
6:27add calendar later after you trust the
6:30output. Let your agent earn higher
6:33responsibility one step at a time. You
6:36know, we don't get promoted easily,
6:38neither should your agent. And by the
6:40way, if you like these kind of
6:42frameworks and systems thinking, I write
6:45a newsletter that goes deep into these
6:47themes every Tuesday. One insight, one
6:50tool, one practice.
6:52Link is below and it's free. All right,
6:55now it's time to move to the second
6:57pillar, C for creativity. Here, you
7:00bring your rough cut, notes, documents,
7:03background, messy thoughts, whatever you
7:06already have. The agent helps assemble
7:09it into something you can judge. You're
7:12still the director. You own the
7:14decision. So let's do a walk-through on
7:17how to build this agent. This time,
7:19you're giving the agent your rough cut
7:22and a target. It is the same prompt
7:24skeleton from the last agent, the job,
7:27the tool, the categories, the output,
7:29the boundary. And for this particular
7:32agent, the output is where you focus the
7:34most because that's where you're going
7:36to define who it's for, how long, what
7:39format, what voice. Okay, so let's use
7:42this creativity agent to make a
7:44PowerPoint presentation. I open Claude
7:46Co-work on my desktop and point it to a
7:50folder on my computer. That folder is
7:53where I drop my rough notes and it's
7:55where the finished deck will be created
7:58as well. I give it the notes and tell it
8:01what I want. Here are my notes for an
8:03idea. I want to pitch this to the CFO.
8:07Build me a short pitch deck. Ask me
8:09questions if you have any gaps in your
8:11understanding. I want 8 to 10 slides and
8:15I want that pitch to last for maybe 15
8:18minutes. It will read my notes. It will
8:20ask me questions if it has any. It will
8:23even fill the gaps with a little
8:25research and build the actual deck.
8:29What comes out is a real PowerPoint file
8:32that I can open and edit in Microsoft
8:35PowerPoint. Now, I can direct. I see
8:38what's working, what's weak. I can tell
8:40the agent what to change.
8:42Make this longer, make this shorter,
8:45make this a three bullet slide. Can we
8:47make the opening slide cleaner? Delete
8:50this slide. Whatever comes to mind. The
8:52agent will help you rebuild. Now, this
8:55deck came out as a real file because
8:57Claude uses something called a skill.
9:00That's just a set of instructions for
9:02doing a specific job. It ships with
9:04skills for PowerPoint, Word, Excel,
9:07PDFs, which is why you get real files.
9:11Now, what if you want to change the look
9:13and feel? Well, you can drop one of the
9:14decks that you like and ask the agent to
9:17match that style. Another cool way of
9:19doing it is that you can even ask the
9:21agent to turn your brand template into a
9:23skill. From that point, the agent will
9:26then always apply the same colors, the
9:28same fonts, the same layout to every
9:31deck and document you'll ever make. Now,
9:33this is just one example of using
9:36skills. You can come up with 100 others.
9:39We're not just learning tools here.
9:41We're learning a new way of thinking.
9:43Now, make sure that the ultimate
9:45direction comes from you. These agents
9:48are your multipliers. If you come in
9:50clear, AI will multiply your clarity. If
9:53you come in confused, well, the AI
9:56multiplies the confusion. And the best
9:58way to create that clarity is to do a
10:00lot of experiments
10:02and figure out what's the best way to
10:04tweak your output and even your agents.
10:07And over time, you'll build a very
10:10strong intuition
10:12>> [music]
10:12>> to get what you want. But what if you're
10:14spending hours trying to understand a
10:17confusing and complex document? What if
10:20you need clarity to even understand
10:22what's going on in that document? That's
10:24our third time-saving pillar. C for
10:28clarity. We've all signed things we
10:31didn't understand. A loan contract, a
10:33car lease, a medical report, an
10:36insurance policy. It's all right there.
10:39It's correct, but it's so confusing
10:42because many of these documents are
10:44written in a language built to protect
10:47those who wrote them. That's where the
10:49clarity agent will save you a ton of
10:51time and hassle. It's your translator,
10:54and that's what we're going to build.
10:56The best way to understand what we're
10:57building is to think about two scopes, a
11:01telescope and a microscope. Sometimes
11:05what you're looking for is scattered
11:07across so many places that you need the
11:10agent to pull it all together. And
11:12sometimes it's the exact opposite.
11:14Sometimes what you're looking for is
11:17buried so deep in a single dense
11:20document that the agent has to dig in,
11:23extract the key ideas, explain the
11:25jargon, flag the risks. That is your
11:28microscope. So, let's go through both of
11:31them and build both ideas. Say I'm about
11:34to sign a deal with a company I don't
11:37know well. So, I can tell the agent,
11:39find out who they are. Find out
11:41everything you can from the internet.
11:43Analyze our past emails with them
11:46because, hey, we've already built our
11:48coordination agent, remember? Find out
11:51anything recent in the news.
11:53And tell me more about this deal that
11:56I'm trying to strike with them. And if I
11:58have uploaded documents, the agent will
12:01search all my documents, all my
12:04connected tools, my Google Drive, and
12:07the web and come back with one picture.
12:10And what if I want to know more about
12:13their product or competitive strategy?
12:16We let the agent do the work for us. And
12:18the prompt could be, now go and do deep
12:21research on their core product from
12:23reliable sources, create a document in
12:26my folder, and cite the sources. Cloud
12:30agents already have web search and deep
12:32research built in. Same for Gemini and
12:35ChatGPT. The interface may look
12:38different, but it's the same idea. Two
12:40phrases I use the most when I work in
12:43telescope mode. Please verify,
12:47and be concise. They can be super
12:49helpful. Sometimes I take the research
12:52output from one AI engine to another one
12:56and make sure I can cross-verify. That
12:58is my advisory board. And there are
13:01three members in it, Claude, Gemini, and
13:04ChatGPT. And sometimes they fight among
13:07themselves, which is great for me. Now
13:09it's time to change our scope. We're
13:11going to use the microscope. Let's say
13:14we have a contract from the same
13:16company. I uploaded the first mistake
13:19most people make is asking, summarize
13:22this contract. A summary makes a
13:24confusing document shorter, sure. But in
13:28legal contracts or long policies, the
13:31devil is always in the details. So
13:34instead, my prompt would be, "Read this
13:38document carefully.
13:40Find the key terms like fee structure
13:42and obligations and deadlines and
13:45exclusions and risks.
13:48Pay attention to anything that's unclear
13:51or anything that makes me liable."
13:54Create a table with five columns. What
13:57the contract says, what it means in
14:00plain English, why it matters, the risk
14:03level, and the questions I should be
14:06asking. Again, that's just one of many
14:09prompts you can give it, but you get the
14:11point. You want to break that document
14:13open, not get a bird's-eye view on it.
14:16That's what agents can do for you. And
14:18now, one more important thing, your
14:22mileage may vary, but I personally don't
14:25like uploading sensitive information
14:28like personal financial data or medical
14:32reports. So, be mindful about all of
14:35that. All right, we're on to the fourth
14:37pillar. This last agent helps you handle
14:40the moment once you're in the room, the
14:42interview, the raise, the pitch. With
14:45some conversations, you only get one
14:47shot. That's why you have to rehearse. I
14:49remember my first job interview. It was
14:52long time ago. And I had no idea what to
14:55do, what to wear. I didn't even know how
14:58to tie a tie.
15:00I don't think I even owned a tie. So, I
15:03asked my roommate if I could borrow his
15:06tie. His name was Mike Green. Awesome
15:09awesome roommate.
15:11He said, "Sure, but do you know how to
15:14wear it?"
15:16And I shook my head and we both cracked
15:18up. So, he got a tie, tied it around his
15:21neck first, loosened it, and handed it
15:24to me and said, "Good luck."
15:26Of course, I walked in wearing a
15:28borrowed tie, but still had the same
15:30anxiety about my first interview. Turns
15:33out that when it comes to interview
15:34anxiety, I'm not alone. In a survey done
15:37by JDP, 93%
15:40of people said they get anxious about
15:42job interviews. That's almost all of us.
15:45But the survey also said something very
15:47useful. One of the best ways to reduce
15:50that anxiety is to practice that
15:52interview out loud. That's the whole
15:54point of the coaching agent. The first
15:57time you say the most important thing,
15:59you should not be in the most important
16:01room. We are going to build an agent
16:03that helps you rehearse. I know CEOs who
16:07rehearse their entire board meeting with
16:10six or seven agents, where each agent is
16:13designed to mimic the personality of a
16:15specific board member.
16:17Yeah. It's that important. We will use
16:20job interview as an example as we build
16:22it, but you can use it for any
16:25professional or personal conversation or
16:27negotiations. Okay, first, I add all the
16:30context, the company background, job
16:33description, my cover letter, my resume,
16:37my position statement, [music] and so
16:39on. Second, we want to tell the agent
16:42who to be. You are the hiring manager
16:45for a senior product role. You are
16:48sharp. You're a little skeptical.
16:51And you've read my resume, so go ahead
16:54and interview me. Ask one question at a
16:56time. And push back on weak answers like
16:59a real interviewer would. And keep track
17:02of our entire interview so we can
17:05analyze and improve my performance. And
17:08here's the move that makes it all real.
17:11Don't type it. Talk to it. Turn on the
17:14voice mode. On your phone. So ChatGPT,
17:17Gemini, Claude, they all let you speak
17:20out loud, and they all can have natural
17:22conversation with you. You can interrupt
17:24it, you can ask clarifying questions.
17:26It's a real conversation. I do this all
17:29the time on my walks. And when I'm done,
17:32I want both the agent and me to break
17:34character. I would say, "Okay, we're
17:37done with the interview. Now, you're my
17:40interview coach. Tell me where I was
17:42weak,
17:43where I fumbled,
17:45what I should have said.
17:47Tell me where I rambled on. Give me
17:50three ways I could have answered that
17:52question better." And once the coaching
17:54session is over, now you can turn up the
17:57heat. Give that agent a persona of a
18:01tougher hiring manager. Maybe you're
18:04applying to a startup, and the CEO will
18:06be the one interviewing you at the end.
18:09Give that persona to the agent. Let the
18:11agent know that it should conduct the
18:14interview in that persona for 45
18:17minutes, and give you the last 15
18:19minutes to ask questions to the CEO.
18:22Because that will happen in real life.
18:25And finally, once you're done with all
18:27your reps,
18:28don't forget to ask. Turn what you've
18:31learned into a one-page prep card, so I
18:34can review it before the interview. Now,
18:37you have something you can use. [music]
18:38Your goal here is not to create a set of
18:42scripted answers, but to build the
18:44capacity to deliver good answers, no
18:47matter what the questions are. Like all
18:50great musicians, rehearsals teach you to
18:53build judgment and taste, so you can
18:56become comfortable improvising on stage.
18:59So, those are the four pillars. You
19:02know, as AI becomes more entangled in
19:06our lives, there's definitely a sense of
19:08fear running under all of this.
19:11According to a recent survey, 70%
19:14of Americans believe AI will shrink job
19:18opportunities. And among Gen Z, it is
19:2181%. Sure, some jobs will be displaced.
19:26There will definitely be some pain
19:28around that.
19:30But, many, many more new jobs will be
19:33created over time because of AI.
19:37We have been here before.
19:40Every age has believed that the machine
19:44would swallow the person.
19:46Charlie Chaplin showed it in Modern
19:48Times. A man literally pulled into the
19:51gears of a factory. And that was made in
19:541936.
19:55If the contest is speed, memory, or
20:00faster output, machines will win every
20:04single time.
20:05But, I think the real contest will be
20:07about the things machines cannot own.
20:11Your attention,
20:13your creativity,
20:15your clarity,
20:17and that one thing that matters most
20:19when you build your career, your life,
20:21your company,
20:22your judgment.
20:25Those are the four pillars we talked
20:27about.
20:29Sure, the machines will keep getting
20:31smarter and faster.
20:34Let them.
20:36Use the machines.
20:38You don't have to become one.
20:42See you next week. Thank you.
20:45And I love you.

About this transcript

This page contains the full transcript of 4 AI Agents To Automate 99% Of Your Life by Sandeep Swadia, generated from the public captions YouTube serves with the video. The transcript has 3,158 words across 480 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.