YT YouTube2Text

YouTube2Text → Transcripts → I built an AI Agent in 43 min to automate my workflows (Zero Coding)

I built an AI Agent in 43 min to automate my workflows (Zero Coding) — Transcript

by Greg Isenberg · 8,692 words · 1,173 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00this episode is with Max the co-founder
0:03and CEO of gum Loop if that name rings a
0:06bell it's because they just raised $9
0:08million they're one of the best-in-class
0:10AI agent and AI automation companies now
0:14this is a software that you can use to
0:17basically automate what a junior
0:19employee does prior to this interview I
0:22have not used it so I reached out to Max
0:25and I said hey can you give us a
0:26tutorial how to automate what a junior
0:29employee does we use gum Loop to do it
0:33uh this is probably one of the most
0:36requested uh episodes gum gum Loop is
0:39super hot right now and I'm excited for
0:42you to learn how to automate more things
0:45in your business this is relevant for
0:48the biggest companies in the world to a
0:49solo preneur enjoy the
0:52[Music]
0:57episode baby
1:01all right so super excited to have Max
1:04here this might be one of the most
1:07requested tools AI tools that people
1:11want me to cover we're talking gum Loop
1:14today we've got the founder co-founder
1:17CEO of gum Loop to literally I mean you
1:21tell us what are we doing today yeah I
1:24think uh uh we're just going to demo gum
1:26Loop I'll show you how it works how to
1:27get started with it honestly demoing is
1:29my my favorite thing to do so I'm super
1:31excited to do it in general and to be on
1:32the Pod I appreciate you having me and
1:36just to be
1:37clear I'm not involved at all in gum
1:40loop at all I just have heard good
1:42things that people are
1:44using gum Loop for a lot of really
1:47interesting automations and could you
1:51tell people Max by the end of this
1:54podcast what will they learn so that
1:56they can implement it in their own
1:58businesses like it's cool that your demo
2:00gum Loop but that's about you what are
2:02people going to get out of this yeah I
2:05think by the end of the Pod you should
2:06be able to build super powerful Ai
2:08workflows and deploy them uh so you'll
2:11have usable tools you can share with
2:13friends that do any number of kind of AI
2:16actions with your data um sounds kind of
2:19vague but uh I can uh show you exactly
2:22how that works beautiful all right well
2:25let's just let's get our hands dirty
2:28sweet I guess to start here a gum Loop
2:31workflow is just a series of steps
2:34basically that you're connecting you're
2:36passing data from one node we call it to
2:39another and uh before you know it you're
2:41kind of able to build these super
2:43powerful AI powered workflows that
2:45connect directly with your data so this
2:48is what we call the Hub these are all of
2:50my
2:51workflows uh and if I were to just
2:54introduce you to the idea of like making
2:56a workflow from
2:57scratch you you're faced with this empty
2:59can canvas and what you do is you add
3:02these nodes onto the canvas so we get
3:04compared a lot to like zappier and make
3:05and other kind of no code automation
3:07tools mainly because we have a lot of
3:09the same Integrations so like if you
3:11wanted to pull data from your slack your
3:13air table your outlook notion Reddit uh
3:17all of these sort of Integrations you
3:20can do that but the the power really
3:22comes in when you connect this data with
3:25AI steps so let me let's say I want to
3:28read from my Gmail inbox I drag this
3:30Gmail node onto the canvas it's reading
3:33from a certain
3:34inbox um and then I want to pair that
3:37with one of our AI steps so we have many
3:40different AI steps they're all powered
3:42by llms the most basic one is ask AI
3:45this is like asking chat GPT a question
3:48except it doesn't just have to be open
3:49AI models you can really plug and play
3:51with any model um even like Loop in your
3:54Azure uh the models you have deployed on
3:58Azure um
4:00and you connect the flow of data just by
4:02connecting nodes like that and I could
4:04write a prompt saying like summarize
4:06this obviously this is not a super
4:07useful
4:09workflow but that is how you connect
4:12data and these AI steps so beyond an Ask
4:15AI node we have a lot more nuanced sort
4:17of AI tooling here so you can do things
4:19like data extraction where I could
4:21specify a schema like I want the amount
4:23and the contents of this is this is like
4:25a receipt or something I can have ai
4:27extract the amount with uh claw 3.5
4:30Sonet and let's say I want to also
4:32specify the
4:33date I now have uh oh cap stop I now
4:37have two pieces of data that I'm
4:39extracting uh that you can use later in
4:42your workflow you can also do like
4:43categorization summarization scoring you
4:45can do video and image analysis really
4:48when you pair this these llm steps with
4:50your data you end up with like fully AI
4:52powered products that would have
4:53normally taken an engineer to build and
4:55that's kind of what we're delivering
4:57with gum loop I can show you quick break
5:01in the Pod to tell you a little bit
5:03about startup Empire so startup Empire
5:06is my private membership where it's a
5:10bunch of people like me like you who
5:13want to build out their startup ideas
5:16now they're looking for content to help
5:18accelerate that they're looking for
5:20potential co-founders they're looking
5:21for uh tutorials from people like me to
5:25come in and tell them how do you do
5:27email marketing how do you build an
5:29audience how do you go viral on Twitter
5:31all these different things that's
5:33exactly what startup Empire is and it's
5:35for people who want to start a startup
5:38but are looking for ideas or it's for
5:40people who have a startup but just
5:42they're not seeing the traction uh that
5:45they need so you can check out the link
5:47to Startup empire.co in the description
5:51I think this workflow this one
5:53here uh is kind of interesting because
5:56this is actually how we generated all of
5:57our Revenue in the early days of of YC
6:00uh so my standard workflow is I I would
6:03wait for people to sign up for the
6:04product and then when I saw their email
6:06I would Sprint to my laptop Google who
6:08they were try to figure out everything I
6:09could about them and then email them if
6:11they were a high value lead we were kind
6:13of shooting for Enterprises to use the
6:15product so that was what we were looking
6:17for but this workflow emulates that
6:19entire process for me so I think this
6:22one also exemplifies two cool features
6:24on gum Loop that make it different from
6:25other
6:26platforms um one is subflows which I'll
6:29get into and the other one is how this
6:31is triggered so the input to this flow
6:33is a user email this example one is like
6:36Tim at HubSpot that's uh imagine that
6:38person signed up for the product but I
6:40don't sit here all day copy pasting in
6:42people's emails this is all running via
6:44web hook so like you can treat any gum
6:46Loop workflow like an API trigger it
6:49from your own product based on an event
6:51um in our case the second someone
6:52presses create an account on gum Loop
6:54their email gets passed into this
6:55workflow and it runs in the
6:57background that email then gets passed
6:59in to this step here which is not
7:03something you can find in the node
7:04Library it's not like an AI step it's
7:05not an integration it is a subflow so
7:08this is an entirely separate workflow
7:10used as a node and it kind of has like
7:13the power of a function if you're a
7:15programmer it's reusable it's clean uh
7:18it's extensible you can give this to
7:19your cooworker they can import it to
7:21their workflow and kind of like benefit
7:22from the work you you
7:23did but in this case this is doing all
7:26the heavy lifting for the research so it
7:28takes the email it gets the domain out
7:30of it it scrapes their website it
7:33summarizes what they do as a company
7:34with Cloud 3 Hau I even went a little
7:37deeper here and I edited the prompt
7:38template under the hood to say like
7:39please just give me a oneliner I want
7:41this to be hyper concise it extracts
7:44their company name this was a later
7:45addition I made because if you try to
7:46get a company name with like Zoom info
7:48or something you'll often get like gum
7:50Loop YC w24 which is not like I can tell
7:53you're Bing if you're uh calling me that
7:56so I I ask AI just to get the the
7:57company name and then I do use those
8:00different Enrichment Services in this
8:01note here so it's doing a waterfall
8:03approach I'm getting industry Revenue
8:05country but you can get like LinkedIn
8:07URL monthly web traffic total funding
8:09number of employees if you
8:11wanted all of that gets funneled into
8:13this little formatter not super
8:15necessary to be honest but it it looks
8:16nice because I'm about to send this into
8:18our company slack so I want it to be
8:19like a nice readable notification I can
8:21open on my
8:22phone and then all of that funnels into
8:25the company slack so like 5 seconds
8:26after someone presses create an account
8:28on the platform we get a ping saying how
8:30much revenue they're making where
8:31they're based a link to their company a
8:32summary what they do um and then it also
8:35drafts an outbound email in my inbox so
8:39it's linked to my Gmail I click save as
8:41draft because I don't want to spam
8:42someone with like a hyper personalized
8:44message right when they sign up uh and
8:46then at the end of the day I used to go
8:47through like maybe we had in the early
8:49days like 20 signups in a day I would
8:51send off the top 10 delete the rest and
8:53and that was my workflow that would get
8:54them on a demo call um so it was kind of
8:57like a semi inbound semi upbound sort of
8:59thing
9:00uh and yeah we built this in the early
9:02days you could definitely extend the
9:04workflow like you could add perplexity
9:06to it you could research their company
9:08way more you could uh do a bunch of like
9:11Google searching you could add them to
9:12your air table enrich your CRM it really
9:15is like a build your own software sort
9:17of platform but in our case like we kept
9:18it simple here uh and I like kind of
9:21keeping this workflow as is because it's
9:22a relic of the past for
9:23us um yeah
9:27so my first reaction to this is this is
9:31probably this is probably the most
9:33beautiful automation
9:35software i' I've seen yet like it just
9:38look it feels a lot more fun and human
9:42to
9:43use um which is awesome like it feels
9:46like it's like hooking the stuff up is
9:48gonna it feels like Lego building blocks
9:50which of you know and it's kind of got
9:53that like figma feel um yeah thank you
9:56so just for that yeah just going from
9:58that like design perspective I love that
10:01I think that something like this like
10:03this is the difference between if you
10:04make $50,000 a year in sales or you
10:07could make 5 million having these tight
10:10systems that just
10:12automatically you know are in your
10:14workflows um it's the it's your unfair
10:17Advantage frankly like you know most
10:20people aren't doing it that's that's the
10:22advantage I think one of the things I'm
10:24thinking about um because I'm like oh
10:28how do I implement this on some me of my
10:30companies is does everyone need
10:34to create one of these workflows
10:36themselves or is there are there like
10:39templates that I can go and
10:41grab yeah so that's a good question you
10:44we often see at companies that there's
10:46like a couple like maybe five AI people
10:49who really love this sort of thing they
10:50build amazing tooling and then everyone
10:52else just uses it um and that's a I'll
10:55show you a few features that really
10:56enable that so like people can actually
10:57use the workflows without even seeing
10:58this crazy flowchart sort of thing um
11:02but also we do have a template directory
11:03and we're going to be launching a
11:05marketplace where people can publish and
11:06even sell their templates so if you
11:08wanted this like crazy enrichment flow
11:10that someone worked hard on you could
11:11play pay a few dollars and just clone it
11:13into your
11:14workspace um but yeah I think that's the
11:17future like being able to share useful
11:20tooling one one thing we like to say
11:22internally is that uh understanding a
11:25problem should be the only prerequisite
11:26to solving it like you shouldn't need to
11:28have an engineering degree to understand
11:30like how to boost sales with automation
11:32you should just need to understand how
11:33to write a great email or what data to
11:35look for in a good customer so that's
11:38kind of what we're we're gunting for and
11:39I think templates are kind of a Band-Aid
11:41over the product being a a little too
11:43confusing at at the beginning if we can
11:45nail this ux problem I don't think
11:47you'll need templates because you can
11:48just get to something amazing in like
11:49five minutes of
11:50tinkering totally and that's why I
11:52wanted to have you on the Pod honestly
11:54is just I wanted to see I wanted to show
11:57people what was possible so that they
11:59can go and just mess around on on you
12:02know on their own because that's that's
12:03the number one way that people are going
12:05to learn realistically is trying things
12:07out and then being like you know what um
12:09I actually like sending the personalized
12:11email right away for whatever reason um
12:15and that that might work for someone and
12:17it just not might work for you or me or
12:19something like that could you uh could
12:21you show something else yeah I have a
12:24bunch of stuff to show oh amazing um I
12:27also built some some workflows that you
12:28could use uh like I stayed at the office
12:31a little later last night and just made
12:32some that are kind of cool but just to
12:34uh show off this and how you run it
12:37because you can run any workfl manually
12:39I realized I haven't uh I haven't shown
12:41how it works so it it breaks down every
12:43single step you can view the inputs and
12:45outputs you can go deeper and
12:46investigate what's happening in the
12:47subflows I just got the slack
12:49notification I have the draft in my
12:51inbox um just wanted to like illustrate
12:54how this actually runs but yeah let's
12:56let's dive into another workflow
13:02here uh this one watches all of your
13:05videos and gives me an analysis of how
13:07you run these interviews we thought that
13:08was kind of fun to make but I think this
13:10one's pretty
13:11cool this exemplifies the value of loop
13:14mode as we call it so this is kind of
13:16how large companies run at scale you can
13:18run automations manually in this kind of
13:20one-off fashion you can trigger them
13:22with events or web hooks or you can like
13:25kick off a run that does thousands of
13:27events at the same time it will Loop
13:28your automation and and be able to run
13:30them at massive scale so this this flow
13:34here uh takes in a YouTube link to your
13:37podcast and it writes a blog post about
13:39it so uh this video here dude I need
13:43this yeah I actually need this I built
13:47it thinking that you would uh enjoy this
13:49so like this is one of I I'm not paying
13:51for premium on this account but this is
13:53one of your your episodes with Samir
13:55call from Colin and Samir yeah let me
13:58just
14:00by the way shockingly this this pod went
14:03didn't get a lot of love but it was like
14:04one of my favorite episodes of yeah I
14:08really enjoyed this
14:09one um so I I I picked that one
14:13specifically because I thought it was
14:14like kind of jam-packed with knowledge
14:17and and tips it wasn't just like a I
14:19think the demo ones maybe are harder to
14:20convey as a blog post but the ones where
14:22they're giving advice and info those are
14:24really great so I think this is uh this
14:27sort of AI content generation workflow
14:29is interesting because you're not like
14:31making content out of thin air you're
14:33just repurposing kind of the hard work
14:35you've already done previously so all
14:37you have to do is paste in this YouTube
14:38link you press run and then the blog
14:40post will be live in like five seconds
14:44uh on ghost which is our CMS it the
14:47thumbnail is a little grainy because I
14:48didn't get the best thumbnail but it's
14:50uh it gives a tldr so like his main
14:52points that he covered and then at each
14:54of those main points it goes into like
14:55exactly what he was saying about it and
14:57then it embeds your video at the bottom
14:59and also links all of his socials I
15:00didn't even prompt it to do this it just
15:02somehow did that um dude are you kidding
15:05me right now this is
15:07amazing yeah this took like under 20
15:10minutes to build I can show you how it
15:12works yeah let's do it um at the
15:15interesting thing I want to show off
15:16just to start is that this workflow that
15:18takes in one YouTube link can be looped
15:20over a thousand YouTube links if you
15:22just use this flow as a subflow link it
15:24to a Google sheet and then run it in
15:26Loop mode which is like a for Loop
15:28basically on gum Loop so you give it a
15:30thousand inputs it will run your
15:31workflow a thousand times and gum Loop
15:33handles the infrastructure of how all of
15:34this runs concurrently all of the rate
15:36limits you just kind of like go get a
15:38coffee while your content's
15:40generated um okay so do you want me to
15:43dive into how how this works I would
15:45love that sweet so the input is a
15:49YouTube link it feeds into this subflow
15:51which is doing basically all of the hard
15:52work I can open the subflow um that
15:56YouTube link that we pass in dynamically
15:58gets passed into this YouTube node that
15:59we have that gets the transcript we also
16:01have a Gemini node that can use like it
16:04can you can do uh uh speech to text if
16:07you wanted to do that to get the
16:08transcript or you can use Gemini to
16:10actually analyze the video itself
16:12there's a couple it's kind of like an
16:13artistic Choice which whichever
16:15direction you want to go um I could have
16:18just fed that transcript into a single
16:20AI prompt saying like write a write a
16:22blog post but I find if you break things
16:24up into bite-size steps the model will
16:26listen to you a lot better you'll get
16:27much higher quality results
16:29so the the beauty of gum Loop is that
16:31you can just do that prompt one time
16:32work really hard on it and then benefit
16:34from that good prompt like thousands of
16:37times if you run this thousands of times
16:39uh so in this case I take the transcript
16:41and I just ask o1 to give me a digest
16:43like take the fluff out because there's
16:45like kind of casual conversation in
16:47between the points give me something
16:49verbose but informative is there a
16:51reason why you chose o1 versus something
16:54else I think for work like this 01 is
16:57just amazing at like the longer form a
17:00critical thinking so if you have a 55
17:02Minute Podcast that's a lot of content
17:04to process I don't want it to get too
17:05distracted so I'm okay with paying more
17:08in terms of uh just so I can get the
17:10best possible
17:11digest and uh the quality of your AI
17:13steps Cascades for sure so like if you
17:16if you if I were to use uh like GPD 3.5
17:19here it would the next step would only
17:21be as good as whatever the previous step
17:22output it totally makes sense the next
17:26thing I do is I just ask it to write a
17:28blog post so you can change the prompt
17:30here super easily uh you'll notice I
17:32said you're writing in Greg's
17:34perspective um I wanted a tldr because I
17:37think that's useful I like that when I'm
17:39a reader and I said like avoid jargon
17:43sometimes models will just be too
17:45verbose and then the last step here is
17:48uh just to format it in HTML so you'll
17:50see there's these like nice headers and
17:52sections that is like AI is exceptional
17:56at that sort of thing and just taking
17:57content and knowing what should be a
17:59title where to put where to make things
18:02bold um I then manually just link the uh
18:06your video at the bottom because I I
18:08think every blog post should have a link
18:10back to the original source and I output
18:12that so that's the content that's going
18:13into our ghost node
18:15here um it also this workflow is picking
18:18a title which is kind of cool so I just
18:20snip like the first thousand words from
18:23the podcast which is normally where
18:24you're like front loading the
18:25information and I have an extract data
18:27node here so this extract data node just
18:30lets you kind of get really structured
18:32content out of the model you can get a
18:34title without it saying like okay here's
18:36your title that's sort of fluff and you
18:38don't have to beg uh chat gbt to to like
18:41omit those fluff words it just gives you
18:43the value so I'm like I want a very
18:45short title your context your job is to
18:48uh pick a title for this video make it
18:50short something
18:53Snappy uh and then that goes into the
18:55blog content and then the thumbnail
18:57thing I just uh
18:59you can get the thumbnail of any video
19:01with this format I just figured I asked
19:04Claude and it told me this is how you
19:05get the thumbnail of a
19:09video Yeah and this is a a template that
19:12I could share like easily I can make
19:14this available to anyone copy the link
19:16to the workbook and then send it over to
19:17you and then you just have it as if you
19:18built it yourself yeah we'll include
19:20that in the show notes in case people
19:23want to like mess around with it I think
19:25this is so important
19:26because this is a way to get to get
19:29thousands of visits SE from SEO on
19:34autopilot to whatever it is you're doing
19:36so one of the things that you can do is
19:40um you know if you selling a product
19:42let's
19:43say um and you can put at the end like
19:47if you
19:48know uh if you like this podcast you
19:52should buy my product or service right
19:54and then a percentage of those people
19:56are going to try are going to buy that
19:58or service or you can just have a lead
20:00magnet you know you can just be like um
20:03you know give me your you know if you
20:05like the call in sere episode you're
20:07going to like uh the 2025 create a
20:11report that we did and then people click
20:13on it they give their email address you
20:15get the email address and then you can
20:16remark it to them it's just a way to get
20:19yeah it's free visits free signups
20:22leads um this is the type of thing that
20:25you just set it on autopilot and then
20:27you let it compound and and good things
20:30happen yeah also where this content goes
20:33is totally configurable like if you
20:35wanted it to go to your Shopify your web
20:37flow into a Google doc into notion like
20:39you just swap out the last step and it
20:41just reconfigures itself you don't need
20:43to like build the ooth integration I'll
20:45also show you in a couple minutes how
20:47you can build your own integration with
20:49AI you just prompt it and it will build
20:51the node for you that you can use on the
20:53platform yeah I was gonna ask you about
20:54that like how does that work it's uh I
20:58think it's pretty cool I'll demo the
20:59full thing I'll build an integration
21:01live okay um and I'll also I want to
21:03show you gum Loop interfaces next
21:05because this is a feature we're super
21:06excited about mainly because kind of
21:09what I mentioned before if you send this
21:10to your cooworker who's never heard of
21:11gum Loop they have a huge task ahead of
21:14them of figuring out what the hell is
21:15going on like they how would they know
21:18to put in a value here or to click run
21:20or any anything so what we do is we
21:24released a feature that lets you just
21:25make an interface on top of this flow
21:28I'll show you a flow that has a a really
21:30nice interface
21:32uh you have 90 you have 97 pages on this
21:36account yeah I I have many accounts um
21:40so this flow you can see it was cloned
21:43389 times because I I put this one on
21:45Twitter and uh it I think it's really
21:49cool um so I'll just start without the
21:52interface and just explain what this
21:54it's kind of an ad marketing workflow
21:56but it's doing what a lot of people on
21:58marketing team are doing in terms of
21:59competitor analysis on a daily basis or
22:01weekly basis so it starts off by
22:04scraping your competitor's Facebook ads
22:06like anything that they're running on uh
22:09Facebook Instagram uh WhatsApp I think
22:11has ads too it's all just kind of one
22:14location where you can view all of the
22:15active ads it then feeds all of those
22:18ads into the subflow uh the subflow is
22:20wrapped in an a shield so that you can
22:22if the ad is like wonky or there's an
22:24issue you can just skip over that error
22:25kind of a more advanced feature but uh
22:28Gemini is watching the video ad and
22:30analyzing the image so it's coming up
22:31with like on an adby ad basis because
22:33it's in Loop mode it's coming up with an
22:35analysis of like what is this company
22:36trying to achieve with this 302
22:39ad um it with all of those analyses even
22:42if it's like 50 active ads it will dump
22:44all of that at the bottom of this very
22:46large o1 prompt that's saying like
22:48you're an expert in ad analysis I want
22:50you to come up with a overall analysis
22:52of this competitor's ad strategy like
22:54what is what are they focusing on what
22:56does it seem like they're they're like
22:58really want to convey with these ads um
23:00it then formats that in HTML and then
23:03sends management an email so on a
23:06scheduled basis like you can ask it to
23:07run this every morning or every Friday
23:09morning at 9:00 a. PSD and you'll get a
23:13uh sorry my cough you'll get an analysis
23:15of the competitor's ad strategy in your
23:18inbox in a really beautiful formatted
23:19way but like I said before if you send
23:22this to someone no idea what's going on
23:23so this gum Loop interface simplifies
23:25things all you need to really do let me
23:27just refresh this I don't overwrite
23:29anything all you need to do is um send
23:33someone a
23:36link sorry send someone this link and
23:40this is what they see it's a like a
23:42really simple impossible to mess up
23:44version of the app you can just uh I'll
23:48use this as an example I wanted to email
23:50me at Max at gum
23:55Loop gosh my mic is covering my keyboard
23:59all
24:00right um and then they don't even need
24:02to know how the workflow is working you
24:04can go see it if you want to you can
24:05click view workflow but I I'm going to
24:07get an analysis of a competitor's ad
24:09strategy in my inbox and just to show
24:11you how you edit an interface you click
24:13here and it's this kind of really simple
24:15drag and drop build your own UI people
24:18on your team will think you're like an
24:19AI engineer who built this crazy web app
24:21but you just like clicked a couple times
24:23and picked some cool
24:26headers I'll show you the uh output in a
24:28second it
24:31arrives this is really cool I can see
24:33myself using this my team to just get
24:37get smarter could I use this like if I
24:40want to build a a SAS product could I
24:43use this and just have gum Loop actually
24:46doing the the work in the back end like
24:49could I make a website and charge people
24:52to to to do to do this 100% we have many
24:57users doing that with uh web hooks so
24:59you can build your workflow like you're
25:01like engineering this crazy API but then
25:05you just uh I realize I'm doxing my uh
25:08we're gonna need to blur some of this
25:09out if that's okay doxing my API key
25:11yeah we'll do we'll we won't dox that we
25:14will blur that out so sweet yeah um but
25:18yeah you can trigger any workflow via
25:20API and then you can pass in Dynamic
25:22Fields so like if you have a cool form
25:23at the bottom of your website that you
25:24wanted to trigger this crazy workflow
25:26and get them an email sent you just copy
25:29paste that web hook definition and uh it
25:31will run whenever you'd like and you can
25:33also set up an alert on gum Loop so that
25:35you know if things go wrong you get
25:36emailed if there's an issue simplifies
25:39the process I don't know if you can do
25:41this but can you show an example of
25:43someone who uses a gum Loop web
25:46hook uh in in their product yeah a lot
25:50of people have them like in internal
25:52dashboards so like you you pay to be to
25:54use the service and then you have access
25:56to like this crazy PDF processor or like
25:58contract analyzer but uh often times
26:01people don't expose them like on the
26:02landing page because you can just like
26:04Run a thousand workflows that will cost
26:06a lot of credits um got it yeah cool I
26:11can show you uh a few more things custom
26:15nodes in particular I think those are
26:16really cool so max I'm loving this
26:21thanks yeah I I love to demo it's like
26:23the best part of my job for
26:25sure um okay let me show you
26:28this one is just a random workflow I
26:30think it's cool every morning at 9:00
26:31a.m. I get a text and an email about my
26:35entire day so it just specifies like the
26:38start and end time for this Google
26:39Calendar reader it researches every
26:41single person I'm meeting so it gets the
26:42attendee list does all that research of
26:44like finding their company scraping like
26:46their revenue monthly web traffic to
26:48kind of size up how big of a customer
26:50they are or how big of a customer they
26:51could be um it feeds all of that into a
26:54prompt saying like hey I'm Max from gum
26:55Loop this is my calendar this is all the
26:57research
26:58uh explain who I'm going to meet why
27:01they might be important to me this is
27:03what I care about it then has another AI
27:05prompt that makes a tldr like a three or
27:07four point form thing that I can just
27:08see on my phone at 9:00 a.m and then I
27:11get a much more thorough email
27:13report um all of this is just scheduled
27:15you can schedule things with natural
27:17language
27:21uh okay you should be able to schedule
27:23things with natural language this is a
27:26the the UI was collapsed for some
27:30reason yeah you you would click new time
27:32trigger here and then you're able to
27:34describe I I know why okay I couldn't
27:37have scheduled it because it was already
27:38scheduled right um but if if you have a
27:40new workflow you can just describe like
27:42every second day of of every 4th month
27:44at 9:00 a.m. and it'll it'll make the
27:45Cron job for you let me uh I just wanted
27:50to go back and show you
27:52this this was the competitor ad analysis
27:54output that just came in two minutes ago
27:57because it took a bit of time to run
27:58over every ad but it talks about new ads
28:02in the last seven
28:03days uh this these are all of the links
28:05to the ads so I can go watch them uh I
28:09you can change the prompt so that it
28:10formats it differently but I just asked
28:11it to like make things spacious use a
28:13bit of color put Emojis focus on the
28:15content type and structure but it's a a
28:18s simple kind of prompt edit away if you
28:19want a totally different analysis of
28:22ads so cool you can also link together
28:26like if you use that as a subflow you
28:27can speci speci ify 40 competitors and
28:29have one comprehensive ad analysis if
28:31that's more interesting it's like having
28:33a junior ad
28:36assistant totally and it takes like an
28:39hour maybe 45 minutes to build like a
28:41completely custom one from scratch and
28:43the cost like what's the cost of like an
28:46email like that uh so it depends it's
28:49kind of like a a function of how many
28:51ads there are how many times you're
28:52running the Gemini queries like B I
28:54think um it might be like 500 gum Loop
28:58credits which doesn't mean anything to
28:59you as I'm saying that but uh way way
29:03way less than someone spending an hour
29:04of their time what's the uh what's the
29:07current you know USD to gum Loop credit
29:10I actually don't know the exact
29:11conversion rate I've been using CAD so
29:14oh no I don't know what yeah what our
29:16conversion is but one of the perks of
29:19gum Loop is that you can actually
29:20provide your own API keys if you wanted
29:22to let's say you already pay for Apollo
29:25you already pay for Gemini you pay for
29:27like five other
29:28you can put your own API keys and then
29:30the cost on gum Loop will basically drop
29:31to zero so we have some really large
29:33customers where kind of like credits
29:34maxing they just put in their API keys
29:37and all of their workflows are like five
29:38credits the cheapest plan comes with 30
29:40credits or 30,000 credits um so they can
29:42like really run at scale without paying
29:44a crazy amount so 30,000 credits it's n
29:50$97 right now yeah right now as of
29:53recording
29:54this um so 500 credits would be like
29:59a162 per
30:01email that sounds about right so we're
30:05talking I mean think about how much
30:07you'd have to pay a junior ad person in
30:10San Francisco where you live um you know
30:14you're talking minimum $80,000 a year
30:18yeah right probably yeah I think that
30:20sounds right um You could also make it
30:22cost a lot less on gum like throw on
30:24your own API key watch the credit cost
30:25drop to 50 and then
30:28run it a thousand times if you want and
30:30it's kind of like
30:32negligible
30:34cool I want to show you custom nodes now
30:37so this is a feature I think is is going
30:40to be one of the the focuses of gum loop
30:43on top of the AI stuff where we're kind
30:45of layering on top of the product so you
30:46can just like get your workflow built
30:48for you which I'll go into
30:50um often times we notice that like a
30:53large customer of ours like instacart or
30:55samsara they they really want either to
30:57inter with their internal data like they
30:59want to uh hit their own end points or
31:02they want some third party that we don't
31:04support on the platform so we always had
31:06to build custom Integrations for them
31:08but we what we like thought would be the
31:10best solution is to let you build your
31:11own integration so you go into the
31:14custom node category and uh let's take
31:16for example built with this was one
31:20request that we got
31:22recently someone wanted this builtwith
31:24integration it tells you like what
31:25websites a uh or what products a a
31:29website is built with and uh I can just
31:32copy the docs
31:34here into this custom node Builder and
31:36we'll get a builtwith node in a couple
31:38seconds no so I'll go create custom node
31:41you'll see I'll call it builtwith
31:42because I made this yesterday the input
31:45we can call it company the output data
31:48you can add as many inputs like kind of
31:49just like build your own integration
31:50here uh just by like choosing the
31:53type you can add a logo I won't do that
31:55for now and then I'll say uh make this
31:59integration please paste in everything
32:03it understands the context of what a gum
32:04Loop integration is what packages it has
32:06access to it's telling you put in your
32:08API key right here because I don't pay
32:10for built with but that's where you
32:11throw it in and then everyone in your
32:13workspace which is kind of like how you
32:14collaborate with team members uh on the
32:16gumlu platform they'll all have access
32:18to your custom nodes so I have like a a
32:21studio gibl node for example they have a
32:23public facing API it'll tell you like
32:25stuff about a movie everyone at gum Loop
32:27can use this just like the gum Loop team
32:29built this custom integration for you
32:31and the people who built it can go back
32:32in the code if they wanted to change it
32:34um I could say like add comments
32:38throughout the
32:42code it already had a ton of comments
32:44but it'll write it again now with my uh
32:47even more kind of comment forward
32:49request um so what we noticed is like a
32:51lot of semi- technical people that
32:53companies uh have just used this
32:57interface to build really complex like
32:59Twitter scrapers like people are
33:00scraping Blue Sky for social media
33:02analysis they're like building
33:04Integrations with their Dev Team without
33:06actually being a Dev uh you can kind of
33:08like help yourself since AI is baked
33:10into the product and this is going to be
33:12a huge focus in the future I I think the
33:14next step is what if you could just like
33:16hit command K on the canvas and describe
33:18the the integration you want it will
33:20know how to name it it'll know what icon
33:22to get the inputs it'll know how to
33:23write the code it'll know how to test it
33:25for you uh and it'll just kind of appear
33:27here here so you can just uh you never
33:30even have to enter this this janky
33:31custom node
33:34Builder it's not even that janky
33:36honestly but I would love a world where
33:38I could just
33:39describe like it would yeah it would
33:41make my life easier for sure yeah I I
33:44think one of the janky parts I mean I'm
33:46very critical of uh I I always think we
33:48can do better which I I think is why we
33:50get to our whole team is very like uh UI
33:53forward so when when things start to
33:55look ugly to us we get really critical
33:56of it and want to improve it totally but
33:58little things like the fact that I
33:59pasted in all of the docs I shouldn't
34:02have to do that I could just link the
34:03docs and we could scrape them uh that
34:05would be a lot easier true a lot more
34:06pleasant um you the fact that I had to
34:08like specify company and output the docs
34:11should tell you what the output should
34:12be they should tell you what the input
34:13should be uh I think that's that sort of
34:15direction is how it could be 10 times
34:16better than what it is right now one of
34:18the last features I wanted to show you
34:19is kind of addressing the same problem
34:21as the interface where if someone builds
34:23a really powerful tool you shouldn't
34:25need to understand how the tool works
34:26you should just be able to experience
34:27the value it provides like you don't
34:29look at the code behind the SAS product
34:31you love you just use that tool or use
34:33that product so we we created a Chrome
34:35extension um this flow looks kind of
34:37crazy but I'll explain what's happening
34:38in a second the main thing I wanted to
34:40point out is that this flow starts with
34:42a Chrome extension node and what that
34:45does is it allows the flow to be
34:48accessible in my Chrome extension here
34:49the gum Loop Chrome
34:51extension um the one I wanted to show
34:53off is a LinkedIn profile researcher it
34:56does everything that I would do as a
34:58recruiter to reach out to like someone
34:59who seems like a good candidate to join
35:01gum Loop um let's go to my co-founder
35:04for example let's say I wanted to
35:06recruit this guy I would go and press
35:08play on this workflow I have pre-built
35:10it's taking the entire content of the
35:12screen and throwing that into the
35:14workflow just like you're kind of copy
35:15pasting it except a lot more seamless uh
35:18it could also screenshot it could look
35:20at the there's a bunch of different
35:21actions that the Chrome extension can do
35:22but in this case it's
35:23scraping and then it feeds all of that
35:26let me actually go to the Run itself it
35:28feeds all of that um through the
35:31workflow which is doing like the
35:33processing and Analysis of the of the
35:36content of the profile so the first
35:37thing it does is it extracts like name
35:38Title Company University location it
35:41summarizes the person's background I
35:42specified that I care about certain
35:44properties in a candidate um like being
35:46a founding engineer or a Founder before
35:48so it it gives me the notable
35:50details it then finds their Twitter and
35:52their GitHub this is kind of like a
35:54cheeky subflow where I Google their name
35:56uh I have a
35:58I I have ai look at the top results and
35:59pick the one that's most likely their
36:01socials it dumps all of that in a Google
36:03sheet so that we can track candidates it
36:06pings my team on slack being like hey
36:07guys look at this new profile go add
36:08them on LinkedIn so they can kind of
36:10feel like they're being reached out to
36:12and then it also finds their email with
36:14Apollo and then uh Engineers like a
36:18really basic message you can really go
36:20crazy on this if you wanted to like use
36:22this in
36:23production most of the people who are
36:24using this type of workflow have tons of
36:26context about like what the company does
36:27and why they're reaching out maybe a
36:29bunch a bunch of example emails and then
36:31it puts all of that in my inbox as a
36:33draft so even if like an intern on our
36:35team finds an interesting candidate they
36:36can get an email in my inbox so that the
36:38founder is reaching out um and this lets
36:41kind of just like everyone benefit from
36:43the tool you're using without looking at
36:45this crazy workflow they just open the
36:46Chrome extension press play uh and
36:49things
36:51happen and Dreams Are Made they are yes
36:56hopefully I think ing the Chrome
36:58extension with the the custom node
37:00Builder just lets you build infinitely
37:02powerful like no code Chrome extensions
37:05which which I find pretty cool what else
37:06have you seen people do a lot of really
37:11nuanced like Outreach is a big one so
37:13like let's say there's a a list of
37:16hotels in this like Niche directory that
37:18you look at and you wanted to email each
37:20one and be like you're ranking third out
37:22of 50 we could make you number one
37:23you're ranking like 402nd out of 50 it
37:25will like with a single click can scrape
37:28the entire page draft 50 emails that are
37:30all relative because it knows their
37:31positioning and then you get like six
37:33hours of personalized outbound done as a
37:35hotelier uh Service Company just done in
37:38a couple seconds um that seems kind of
37:41random but that's like the the first one
37:42that comes to mind it's all about like
37:44imitating the really Niche workflow that
37:46you do every day and just building SAS
37:48for it because I don't think any
37:49vertical SAS startup would ever build
37:51that but now that you can just use Legos
37:52and piece it together you get your own
37:54sess yeah I I was having a car ation
37:58recently with a an executive of a a
38:02large company I do a billion in IA every
38:04single year and he was walking me
38:08through uh basically how this company
38:10works like how the sausage is made and I
38:12was shocked to hear how much of his
38:15company was operating manually like how
38:19few automations he was actually doing in
38:22a world where things like this exist um
38:25and this is like a you know uh Tech
38:28forward company this isn't like I'm not
38:30talking about a company that's been
38:31around for a hundred years and I think
38:34that if you're a big company like that
38:36to a small company to a company of one I
38:39think a good exercise is just writing
38:42down what are the things I'm doing
38:44manually and thinking about how can I
38:46auto you know what can I automate being
38:48in business is a game of of unfair
38:52advantages it's a game
38:54of uh creating the best possible
38:56customer experience and that means it's
38:59always about how do you save time as
39:03Founders and executive
39:05teams and also you know so that we can
39:10work on our like high leverage stuff the
39:11stuff that we think that actually could
39:13move you know can move the needle while
39:15automating away a lot of the stuff that
39:18um is repetitive frankly yeah I like to
39:22say that like when people ask me what
39:24they should automate for their
39:25businesses I like to say that if you can
39:27list it as a a list of steps like for an
39:29intern you would hand off a little
39:30sticky note being like you do these 15
39:32things in a row and that's the entire
39:33workflow then you can 100% automate it
39:36and I think there's like thousands of
39:38those workflows in every business
39:39there's opportunity to automate
39:41basically everywhere it's just about
39:42like understanding what tools can get
39:43you there and understanding the problem
39:45itself to to solve one huge thing we
39:48were noticing was like an SDR or a
39:50marketing person or a growth person at a
39:52company would deeply understand a
39:54problem but they have to like break that
39:56down for an engineering team they have
39:58to like find headcount and budget and
40:00then they'll get some like half baked
40:02solution that is not exactly what they
40:03had in mind because there's kind of this
40:04broken telephone sort of situation going
40:06on but if you give people the tools to
40:09solve their own problems you get just
40:10like hyper custom exactly what they
40:12wanted work blows that that work out of
40:14the box and that's kind of what we've
40:16been doing as a startup so like we I
40:18like to think we we're super efficient
40:20because we automate basically all
40:21business processes at the company with
40:23gum Loop uh that's why I think the
40:25product gets better because we dog food
40:26it um and people ask us like how big the
40:28team is normally and when I tell them we
40:30were two people and two interns up until
40:33just recently they're kind of mind blown
40:35because we seem like we're a larger
40:37company but it's because we can just
40:39like automate the boring stuff and focus
40:40on engineering product uh marketing like
40:44the things that that we we think we're
40:46good at and that matter yeah absolutely
40:49uh I also think that when you compare
40:52using something like gum Loop
40:54to you know sometimes working with human
40:58beings you're you're you know like for
41:01example the first flow that you showed
41:03around sending emails that were so data
41:05enriched a lot of those emails probably
41:08if it was a human behind it it probably
41:10wouldn't be as personalized um or have
41:15as much data um now I know someone in
41:17the comment section is going to be like
41:19yeah but it would be more personalized
41:20it's a human being um and I hear you
41:23like there there is by the way I'm not
41:25saying that like sales needs to be done
41:27automated I'm just saying that I'd
41:29rather have that person you know H you
41:32know in the first video call or in an
41:34IRL meeting than
41:36sending like essentially automated sales
41:40email so U Max I want I want to give you
41:43a just I just want to thank you for for
41:47taking us around gum Loop playing with
41:49it um what should people do if I mean do
41:55you want yeah I guess sign up to gum
41:56Loop and play with it yeah uh we're like
42:00super easy to sign up with totally free
42:02to use at the beginning like you don't
42:03no credit card needed to play around I
42:05think that's like the most honest way to
42:07build a product like if you can't hide
42:08behind a sign up flow or like a book a
42:10demo button people can actually see if
42:12they like it or not which holds us
42:13pretty accountable so yeah go to gum
42:15Loop sign up for free if you do the
42:17tutorial you get a thousand free credits
42:19so you'll actually have like quite a lot
42:21of space to just play around um and we
42:23have a community we're building a slack
42:25Community now so like we'll have a we
42:27have Discord but we'll have a slack
42:28Community where you can ask questions
42:30get help from experts and then pretty
42:32soon you'll be able to just sell your
42:33workflows share them with
42:35others if you have any questions just uh
42:38I guess email us or yeah join the
42:40community yeah and I think you know my
42:41perspective is if you're not if if
42:44you're not automating stuff in 2025 like
42:48you should
42:50be um so like play with the tools like
42:54play with gum Loop you know play with
42:56the other tools that are there um find
42:58what works for you um and it's a there
43:03is like a learning curve in terms of
43:04like okay now I understand what I could
43:06automate and from that you can just
43:08layer on more and more automation so um
43:11thanks again for coming on Max uh you're
43:14a legend and uh I'll see I'll see you
43:17around yeah thanks for having me bye
43:20everyone
43:21[Music]
43:29it's time baby

About this transcript

This page contains the full transcript of I built an AI Agent in 43 min to automate my workflows (Zero Coding) by Greg Isenberg, generated from the public captions YouTube serves with the video. The transcript has 8,692 words across 1,173 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.