Latest / The Koerner Office - Business Ideas and Deep Dives with Chris Koerner / How to Build Your Next Business on Top of this New AI Tool That’s Blowing Up | Ep. # 153
Transcript
- 0:00Oh my gosh. I mean, every single one of
- 0:02these flows is a business. Pay me $300 a month, I'll send
- 0:05you an e-mail every Friday with your top three competitors fully
- 0:08automated. And as far as your customers
- 0:10know, you're spending four hours a week doing this for them.
- 0:13We actually hired one of our experts because he was just so
- 0:15good. I was like, what if you could do
- 0:16this for our largest enterprise accounts?
- 0:18All he had to do was like learn this drag and drop tool and now
- 0:20he has a new career. You can build software that is
- 0:22immensely valuable without needing to be an engineer.
- 0:24Almost everything that people really build on government and
- 0:26find valuable, they could have never done on Zapier.
- 0:28Use AI and. Be more efficient.
- 0:30If you're not using AI, you're going to fall behind.
- 0:35All right, you're going to love today's episode.
- 0:36You may have heard of this company out of Silicon Valley
- 0:39called Gumloop. They are AY combinator backed
- 0:42company blowing up like crazy. If you've heard of Zapier, I
- 0:45talk about Zapier a lot. It's an automation platform
- 0:48that's easy for anyone to use. These guys are the Zapier for
- 0:51this new AI age. It's like Zapier but very easy
- 0:54to use. I had the Co founder on to talk
- 0:56about how people are using gum loop in interesting ways, how
- 0:59people are making money with gum loop, agencies that can be built
- 1:02on top of gum loop, how to become a gum loop expert so you
- 1:05could charge money to implement gum loop.
- 1:07And I just want you to know upfront this is not a sponsored
- 1:10video or podcast. I was paid $0.00 to do this.
- 1:14But the way I look at it is, what if we could have talked to
- 1:17the Zapier founder in year 2 or year one, and you could have got
- 1:20ahead of that curve and started an agency on top of Zapier.
- 1:23That would have been amazing, right?
- 1:24Well, I look at Gum Loop as the next Zapier.
- 1:26And so anyone listening to this will have a leg up to anyone
- 1:29else in the future that wants to build an agency on top of Gum
- 1:31Loop. But if you do want to support
- 1:33this podcast, you can use my affiliate link.
- 1:35You'll get a discount. I'll get a little kickback.
- 1:38It's LinkedIn, the show notes. You're going to love it.
- 1:39Enjoy. How would you explain Gum Loop
- 1:42in one sentence to someone that has no idea what it is?
- 1:44We're an automation platform where you you drag and drop,
- 1:47build really powerful AI workflows without needing to be
- 1:50an engineer. You can just play around with
- 1:52these Lego blocks and deploy super useful tools for your
- 1:54team. OK, here's an analogy I would
- 1:57use. Tell me if this is accurate or
- 1:58not. But like Instagram was like
- 2:00Facebook except pictures only and mobile first, right?
- 2:04And this kind of reminds me of Zapier except AI first.
- 2:08Like AI front and foremost, whereas Zapier is like you can
- 2:12use AI, you can integrate open AI API, but it's not like AI
- 2:16first, it's automation first. Is that pretty accurate?
- 2:19I think that's accurate. In the early days people would
- 2:21describe us as an AI first Zapier on steroids.
- 2:24So it it's the AI part is, is a big component of gum loop.
- 2:29But also just the complexity of the sorts of things you could
- 2:31build are very different. So what I've noticed is most of
- 2:35the things are almost everything that people really build on Gum
- 2:38Loop and find valuable they could have never done on Zapier.
- 2:40It's kind of like expanding what is possible instead of replacing
- 2:43them. If someone just wants to connect
- 2:45to AP is and get a notification or something.
- 2:48I would say totally use Zapier. That's that's their bread and
- 2:50butter. Much simpler, probably cheaper
- 2:52for the task, but for anything else more complex you just come
- 2:56look for sure. What's a good example of like a
- 2:58common flow that is just not possible on on Zapier?
- 3:02I can give some examples of the ones that we use internally.
- 3:04So like every time someone signs up for our products, we do a ton
- 3:09of like AI web research to figure out everything we can
- 3:12about them. Like every signal that would be
- 3:13valuable in terms of scoring that lead.
- 3:15We run it through a bunch of AI categorization steps that apply
- 3:18like our reasoning to what is this account a key account that
- 3:22we should be emailing? It drafts A personalized e-mail
- 3:24and it puts it in my e-mail inbox and I'll be able to send
- 3:26those out at the end of the day. But there's a lot of nuance,
- 3:29like 4 different models being used, tons of different
- 3:32endpoints being pulled from to enrich data to scrape the web to
- 3:35look at their website. That's one workflow that I use
- 3:39every single day or hundreds of times a day.
- 3:41Every time someone talks to our customer support chat bot called
- 3:43Gummy, they normally are curious about something or they can't
- 3:46figure something out. It would be super useful if I
- 3:49could go and read every chat, but I don't have the hours and
- 3:51they do that. We're doing like 60,000 messages
- 3:53a week on our chat bot. So I have a gum loop workflow
- 3:56that runs at 5:00 PM every day. It reads every single thread.
- 4:00So every conversation with every customer tells me what they were
- 4:02confused about. And then it with all of those
- 4:05summaries it makes one large report that tells me like the
- 4:08top five things with the substantiated evidence.
- 4:10Like this e-mail was having a ton of trouble with this sorted
- 4:14from most to least important. So we kind of have this like
- 4:16product manager report that is totally autonomously done
- 4:20without needing an engineer or to waste any hours doing that.
- 4:23You're like passively collecting feedback on your product with an
- 4:26agent, basically that you built. Exactly.
- 4:29And it's it's using our own product to help us figure out
- 4:32how to improve our own product. It's a nice little tight loop of
- 4:34feedback. You're eating your own dog food.
- 4:37Yeah, we eat a lot of it. I feel like people throw around
- 4:41the word agent a lot. Is this is would you call gum
- 4:44loop a proper agent builder or is it agent like?
- 4:48So we make AI powered workflows at the moment.
- 4:52I think the term agent was super.
- 4:54It's a big buzz term two years ago when the first agent
- 4:58frameworks came out. Those are definitely much more
- 5:01autonomous and inherently unreliable and more expensive.
- 5:05In the in like the previous state, how agents existed, you
- 5:08kind of had to cross your fingers and hope that the agent
- 5:10would do something useful because it just had so much
- 5:11autonomy and there was so much AI being used every step.
- 5:14I think we're getting to a point now where agents are becoming
- 5:16useful, but we've taken the approach of instead of promising
- 5:19people the world and then underwhelming them and losing
- 5:22their trust, which I think is what a lot of these open source
- 5:25agent frameworks did. Like everyone tried it and is
- 5:27kind of scorn because they spent six hours getting nothing except
- 5:31for making a really big open AI bill.
- 5:33We promised very little in the early days.
- 5:35We're like, you can do this step and then you can do another step
- 5:37after it. And it's very rigid and defined.
- 5:40And we're kind of continuously adding these agentic features
- 5:43like if you wanted to split a path of your workflow and have
- 5:46an AI step decide what path to pursue, that is somewhat
- 5:49agentic, but it's pretty constrained and rigid.
- 5:52I think one thing that direction we will be heading in is if you
- 5:56want to make agents useful, they need to have useful tools, which
- 5:58are essentially just work flows in their little toolbox.
- 6:01So there's a layer on top of what we're building that can be
- 6:03more agentic. But fundamentally we we are
- 6:06making AI powered work flows, which I think is the shortest
- 6:08path to just extreme extremely obvious business value.
- 6:12Yeah, So if you come at customers and say we're
- 6:14basically an AI at first AI enabled Zapier and then they
- 6:19they sign up, they log in, they start playing around and they
- 6:22learn on their own. Oh my gosh, Like this has a lot
- 6:24of agentic features. I can I can make an agent with
- 6:26this like they're, they're super users and they're super
- 6:29evangelist of the product. Whereas if you come out and
- 6:31promising agentic features, maybe they're disappointed and
- 6:34it's a net loss in the. End.
- 6:36I think people who. Who are get super excited by
- 6:39agents specifically often think like it'll abstract away all of
- 6:42the hard work of solving or like automating a problem, like it'll
- 6:46just figure things out. But I think the nuance of really
- 6:49automating something well is kind of like encoding what you
- 6:53understand to be like true whatever you deeply understand
- 6:57some workflow that takes a bit of like you're kind of building
- 7:00this workflow with all of your expertise in mind and encoding
- 7:03that into the workflow. If you just ask an agent to go
- 7:05write a sales e-mail, it's going to write some slop, you know,
- 7:08And then technically it did the work for you, but it's not that
- 7:10valuable. So our AI workflows are kind of
- 7:13like assisting the human, making someone 10 times more
- 7:15productive, but not just like hand waving away all of the
- 7:18complexity of how you do work. I feel like a lot of the
- 7:21problems that we're having with AI today is that everyone knows
- 7:24that AGI is not here yet, but they treat it as if it is,
- 7:28right. And I think you need to treat AI
- 7:30like an employee, like you onboard a new employee and
- 7:32you're not going to tell them on day one write this new sales
- 7:34e-mail for me when they have no idea like what your branding is
- 7:37like, how you like to speak to the customer, how many automated
- 7:41follow-ups you want, if any. Like you're just setting that
- 7:43employee up for failure and you're setting the the AI up for
- 7:47failure unless you you bring it something right?
- 7:50Like this is what I'm doing on a daily basis.
- 7:52It takes me about 30 minutes a day.
- 7:53I've done it 1000 times. I'm sick of doing it.
- 7:55I know exactly how it's done inside and out.
- 7:58Do it for me. You show it exactly what to do.
- 8:00You get, you know where it's going to break because it breaks
- 8:02for you sometimes. Would you agree that that's
- 8:05where people have the most success when they kind of it
- 8:08brings? Yeah, we'd like to say that if
- 8:10you can define something as a list of steps, it's kind of ripe
- 8:13for automation. That if you have a workflow that
- 8:15you can't actually break down into a list of steps, or it gets
- 8:19into a bunch of conditionals like, oh, if it's not here, go
- 8:21check there, give them a phone call.
- 8:23If like there's a bunch of random edge cases, then it's not
- 8:26really great to automate. But normally a workflow that you
- 8:29understand that so deeply that you can write it in like a point
- 8:31form list, give it to an intern and have them perform it
- 8:33accurately. That is exactly the sort of
- 8:36thing you should be automating. And the context you're talking
- 8:38about is, is like the kind of the biggest pitfall of people
- 8:41who approach agents or automation.
- 8:43They say, like, write me a sales e-mail with no context.
- 8:46And if you would have given that to your new hire who started
- 8:47today with no information about the company, they would write an
- 8:50equally bad e-mail. You know, there's unless they
- 8:53were agentic and then went and found the information in your in
- 8:56your Google Docs and then figured out the branding from
- 8:59your website. That's definitely like the
- 9:00direction AI will end up in this sort of like totally agentic
- 9:04software. But I think the path there is
- 9:06work flows and tooling around how AI does work.
- 9:09Yeah, OK. Now I want to know like the 8020
- 9:13of how people are using gum loop.
- 9:14Like that's one of my favorite things to do on this podcast is
- 9:17to talk to the people behind the scenes and say how are people
- 9:21making money with this? Like what's the one customer
- 9:23that comes to mind that's really using your tool in an
- 9:26interesting way that could other people could benefit from,
- 9:29right? It would benefit you because
- 9:30you'd get more customers and it would benefit them because they
- 9:32can make money or they can save money with it.
- 9:34So what's the 8020 like? What are most people using it
- 9:37for? And then one of those really
- 9:38interesting edge cases. That's a loaded question, but
- 9:41you just take one at a. Time.
- 9:43So people ask this is all the time.
- 9:45And it's I don't have a great answer for what the most common
- 9:47use case is because I think gum loop is more comparable to
- 9:50Python for example, than it is to a normal vertical SAS that
- 9:53has like 5 features or something.
- 9:55So I mean, when you ask yourself like, what are people doing with
- 9:57Python? They're building like crazy
- 9:59things and each one neatier than the next.
- 10:02There's definitely like local maxima that we see.
- 10:04So there's tons of like web scraping, outbound sales
- 10:07automation, CRM related enrichments.
- 10:10Document processing is a huge one, like general admin tasks
- 10:14like preparation, data labeling and categorization, lead scoring
- 10:18there. There's kind of like patterns we
- 10:20see. Content generation is another
- 10:21really big one. But when we like our most
- 10:24successful customers are doing everything.
- 10:26So like Instacart has 1000 seats on Gum Loop out of the 3000
- 10:30employees and it's almost evenly spread across 10 departments.
- 10:34So there's not one killer use case.
- 10:35It's like 150 really solid automations and they're all kind
- 10:38of empowered to automate work. Almost none of them or I think
- 10:41like 18% of them are engineers. When we did the breakdown, we
- 10:44did the breakdown with a gum Loop workflow.
- 10:46Actually, we loaded the entire user list into a research
- 10:48workflow that found their role and then categorize them.
- 10:50Yeah, that's our our favorite type of customer because like
- 10:52big scale, high volume, you can automate so many things I've
- 10:55never even heard of. And that's like really the power
- 10:58of this platform. 18% of their employees are engineers and 33%
- 11:02of their employees have our active Gumloop users. 18% of the
- 11:06active Gumloop users are engineers.
- 11:08When we do, oh wow. That's even more impressive.
- 11:11Yeah, I think engineers appreciate just how much easier
- 11:13we make auth and security. Like you don't need to spin up a
- 11:16little side project and then deal with compliance and like a
- 11:19lot apps and and all of that stuff.
- 11:21Gumloop makes it super super easy and engineers pick it up
- 11:24like so easily. It's like the fundamentals of
- 11:27the same for loops and thinking in systems passing inputs to
- 11:30outputs. Let's talk about like a local
- 11:32business owner. Let's say I own a plumbing
- 11:34business in the middle of the country.
- 11:36I've got several crews, $10 million in revenue.
- 11:39What's a normal use case for this?
- 11:41It really depends what they're struggling with.
- 11:43So sometimes people are getting tons of orders via PDF emailed
- 11:48to them and they have someone sitting in an office, like
- 11:51extracting the order number, the amounts and the address and
- 11:55putting that into a Google Sheet that is a gum loop workflow that
- 11:57could be set up in like 30 seconds.
- 11:59It's super easy. They might have a list of leads
- 12:01that they're buying from some sourcing company, but people
- 12:03they should be reaching out to, but they don't.
- 12:05They know in their head what a good lead looks like.
- 12:08It's like they have these five signals or sometimes they have
- 12:10this, which means they're really valuable or you shouldn't reach
- 12:13out to them. If they have that, they do a
- 12:14bunch of enrichment on the platform with our AI scraping,
- 12:17AI web research nodes, and then they have a categorization step
- 12:20and then they press run their whole leads list gets processed
- 12:22and categorized and then sorted by the most valuable.
- 12:25They save all of that time. And then if they want to go
- 12:27above and beyond and like generate the outbound or the
- 12:29messaging or like the call, the phone call transcript.
- 12:32So sometimes people will make like a research report for each
- 12:35qualified lead and then have that dumped out as a Google doc.
- 12:37So when they're actually making the call, they have all this
- 12:39ammunition for like the kind of like a research report on who
- 12:42they're about to speak to. You can really go down the
- 12:44rabbit hole like the the more niche the better to be honest.
- 12:47Some loop is just like the infrastructure that lets you
- 12:49convert what you understand to be a workflow into a functional
- 12:52tool that's deployed. Yeah, I'm thinking like if I'm
- 12:54a, if I'm a local handyman or I own a handyman business and my
- 12:57best customer is an investor because they're going to use me
- 13:00every month. Or a property manager, they're
- 13:03going to use me every day or every week or every month.
- 13:05And I have a, a form that they fill out on the website with
- 13:08just a radio button that says, are you a homeowner or an
- 13:11investor, right. And I just kind of know
- 13:13instinctively, if I get an e-mail that says investor, I
- 13:16want to call them more quickly. Whereas if they're a homeowner,
- 13:19only 20% close rate, it might be $100 job.
- 13:21It's not really worth it. Gumloop could help.
- 13:23Like I want to get a text on my phone marked as urgent when
- 13:27that's an investor. And I just want to, you know,
- 13:30one click I want to call them. Whereas if it's a homeowner, put
- 13:32them in my CRM, put them in an automation to send them emails
- 13:35because it's not worth my time to call.
- 13:37Is that kind of a decent use case?
- 13:39Yeah, that's that's a good, nice and simple one for sure.
- 13:41It'd be really to set up, we have a texting node.
- 13:43I have a bunch of text notifications that get sent to
- 13:45me when important things happen. Yeah, totally doable.
- 13:48And then you can expand upon that.
- 13:50The text could include all of the background research and the
- 13:53person who submitted the forms, Twitter account and their
- 13:55LinkedIn and their company website.
- 13:57Like you can really go deeper with it.
- 13:59I think that's a big problem that people have is they hear
- 14:01about these cool tools and they're like, this is amazing.
- 14:03And then they log in and they're like, what do I do?
- 14:06You know, like what? All right, what can I do?
- 14:08I wonder if this is an interesting framework where
- 14:11whatever you're doing, you know, control find control C control
- 14:15V, whatever you're, you know, control finding, copying or
- 14:18pasting, there's probably an automation to be found there.
- 14:20Like we have a, a GPT wrapper in our holding company that helps
- 14:24skilled nursing facilities bill for more codes.
- 14:27And so they'll on board a, a new client and then upload thousands
- 14:31of pages of PDFs and medical records.
- 14:33And for the last 100 years, what the, the, the medical billing
- 14:37manager does is they do control, find Alzheimer's, control, find
- 14:41diabetes control fine. And they miss stuff.
- 14:44And so there's thousands of dollars a month that they could
- 14:46be billing for that they're not. And so we just built a wrapper
- 14:50that you upload all the PDFs and it tells them and it saves them
- 14:53thousands of dollars a month, right.
- 14:55So that just kind of made me think, control fine, control
- 14:57fine. That's inefficient.
- 14:58That's repetitive, right? How could we automate that?
- 15:01There's definitely little signals like that throughout
- 15:03your work day where you should be thinking in terms of
- 15:06automation. I think that's a good one.
- 15:07Like the control fine. But also just whenever you're
- 15:12applying reasoning to like a really repetitive thing, that's
- 15:15really where LLM shine. Like, even if it's a little more
- 15:18nuanced, like you're seeing emails coming into your inbox
- 15:22and you know what an e-mail you should be forwarding to your
- 15:24support team looks like there's some signals, like it's, it's
- 15:27clearly from a customer. They seem, yeah, they seem
- 15:29upset. Maybe those are the only ones
- 15:31you you forward, but you go through 800 emails a day because
- 15:35your inbox is flooded. That's the sort of thing that
- 15:37you could set up in like 5 minutes with GPD 4-O Mini, just
- 15:40doing the same reasoning, doing the same work for you.
- 15:42You only have to set it up one time.
- 15:43And then you have this like totally autonomous support
- 15:46e-mail bot that that works for you.
- 15:48Yeah, and it can analyze sentiment, right?
- 15:50Like if negative, exceptional. Yeah.
- 15:54Yeah, it's, it's, it's really tremendous at just reasoning.
- 15:57So like categorizing into negative, you can even put it
- 15:59into buckets like, oh, this customer's complaining about
- 16:02this specific issue and I know who to route that to.
- 16:05That sort of even more nuanced reason, beyond negative is, is
- 16:08something that you can set up so quickly.
- 16:09You mentioned that like at 5:00 every night you push all the
- 16:12emails through. Are you or your customers using
- 16:15any e-mail automations that they fully trust to just fully send
- 16:19without any human interaction? If So, what do those look like?
- 16:23Yeah, I do that sometimes, but rarely.
- 16:25I normally like to be in the loop, so I always save emails as
- 16:29drafts and then I'm the ones sending it off.
- 16:31The only ones I fully blast off are like, let's say every every
- 16:35week. We like monitor the really high
- 16:36value signups on Gunloop that customers that look like they,
- 16:39their companies could be so much more efficient with the tool.
- 16:41I'll blast out like a template and e-mail.
- 16:43That's to the point. It's just my calendar basically
- 16:46like saying I'll help you out, but let's talk next week.
- 16:48That I trust because there's there's not much nuance there.
- 16:51But if it's a really thoughtful, like researched e-mail with all
- 16:53these different hooks and personalizations, I'll review
- 16:56that sometimes, like there's a customer within that list that
- 16:58you're personalizing for that's just so high value, you want to
- 17:01throw in like an extra something.
- 17:03Yeah. But yeah, review it, it just
- 17:05gives you more Peace of Mind. I think that's the you'd never
- 17:07want to be someone spamming or like flooding someone's inbox
- 17:10with more spam that needs to be processed by AI itself to be
- 17:13deleted. So when someone signs up as a
- 17:15new user, what kind of research are your agents doing or your
- 17:20automations doing? Are are you just taking their
- 17:21e-mail and then running it against LinkedIn databases or
- 17:25what does that look like? So we take their e-mail and then
- 17:29if if it's able to identify their full name in their e-mail,
- 17:32then we have a a search that goes and looks up like their
- 17:35full name, their company name that they signed up with.
- 17:38It's only we only do it for non Gmail, sort of like corporate
- 17:40emails. And then we find their LinkedIn
- 17:42profile, their title, we'll find the company's current like kind
- 17:47of fundraising stage, whether they raise venture capital, if
- 17:49they're like a growth seed or whatever stage fund or a
- 17:52company, we'll get a one line blurb of what they do, the
- 17:55industry that they're in, the city that they're located in.
- 17:58We'll stitch that all together and then maybe enrich it with
- 18:00even more contact information. Like there's a bunch of data
- 18:02enrichment providers that we support on gum Loop to get the
- 18:04more niche signals like exact headcount.
- 18:07Or their monthly web traffic for example, like extra signals that
- 18:10you wouldn't normally get just by scouring the web.
- 18:13But like linking all of that publicly available data plus
- 18:16like the data repositories together into one classification
- 18:19step and then running that through the LLM step is where
- 18:21the magic is. Because it kind of it's like it
- 18:24did all the work in half a second and then just has to
- 18:26reason around the data that it's great.
- 18:28And then are you assigning like a score to them?
- 18:30I tend to do a categorization instead of scoring because we
- 18:34have a scoring node that goes from like zero to 100.
- 18:36But whenever the AI is uncertain, it'll just give like
- 18:39a mediocre score. And I want some certainty in
- 18:42this. So I like doing like high value,
- 18:44extremely high value normal and low value lead.
- 18:47The more discreet buckets LMS are just much better at like
- 18:50placing things in those sort of much more clear holes.
- 18:53Do you know of any customers building businesses on top of
- 18:57gum loop or agencies or AI implementation agencies?
- 18:59Tell me. Tell me what you've seen there.
- 19:01One of our first customers was this healthcare company in
- 19:04Australia that was processing thousands of research papers for
- 19:09academic institutions. And all of it was powered via
- 19:12like a really complex gum loop workflow.
- 19:14It was called agent Hub at the time.
- 19:16So an agent hub workflow, but same, same product.
- 19:19We're doing things I did literally did not understand.
- 19:21I didn't understand the value of it, but I knew it was valuable
- 19:23because they were like reselling.
- 19:25From what I understood it was like at scale, extracting like
- 19:27the insights for all these papers and doing this cross
- 19:30analysis to determine like what the academic institution should
- 19:33be investing in or what their weak spots are based on the
- 19:35research that they're like PhD and Masters students were doing.
- 19:39But that service would have been hundreds or thousands of hours
- 19:41of human work. But this person just kind of
- 19:44cobbled together a workflow and then built a business on top of
- 19:47that. Interest and they're still
- 19:48going. Yeah, they're still one of our
- 19:50favorite customers. What do you think is next for
- 19:53Gumloop that people should be excited about?
- 19:56We're working on our agent feature.
- 19:58So not an agent in the normal sense of like you're building
- 20:01agents that'll come at some point.
- 20:03But we are making a copilot basically in the product.
- 20:05So right now we're we have like really good growth and things
- 20:08are working, but we could be growing way faster if the
- 20:10learning curve was flatter. Like there's so many people who
- 20:13come to the product and had they have a desperate need.
- 20:15They really want to automate something, but they can't figure
- 20:17it out because it takes a bit of time, maybe like half an hour of
- 20:20thinking around with things to figure stuff out and people are
- 20:22busy. So this copilot feature will let
- 20:24you kind of go from natural language to.
- 20:27This sort of bolts like experience or B0 like experience
- 20:30of building for you and that will change everything maybe.
- 20:35So you just tell it what you want, and instead of telling you
- 20:38how to build it, it will build it for you.
- 20:40Exactly. It'll build it with you.
- 20:42I think that's the main distinction.
- 20:44You get to guide it, steer it in the right direction, work with
- 20:47it, and it has. It is all knowing.
- 20:49It knows everything about gum loop and how it works.
- 20:50You don't have to read our docs, you don't have to watch a video.
- 20:53It's already done that for you. Have you ever thought of a
- 20:55feature where people could just like I would love to go into one
- 20:59of my employees and say, hey, I want you to loom everything you
- 21:02do today, like turn off your front facing camera, hit record
- 21:05on loom. I just want to record everything
- 21:07that happens on your screen. And this is to empower you not
- 21:10to take your job. And then like have gum loop or
- 21:13something, manus, whatever. Just watch that and say, all
- 21:17right, oh, I know exactly what you need.
- 21:18You keep doing this over and over and it's something that
- 21:21that employee just might not even think to tell you.
- 21:23It's just second nature to them. Have you thought about something
- 21:26like that? Or at least part like enabling
- 21:28that more easily so people could have more ideas of how to use it
- 21:31I. Think there's a lot of companies
- 21:33that are doing that as like a Chrome extension or as a sort of
- 21:36screen recording automation tool.
- 21:39I think that is another sort of product that has a big promise
- 21:42and might under deliver greatly, which we're kind of wary of
- 21:46because there's only so much user trust you can burn before
- 21:50you like destroy the viability of the company.
- 21:53So I don't know when I think about the tasks I'm doing
- 21:55everyday, like I'm actually thinking a ton in a way that is
- 21:58not visible on my screen when I'm, when I'm about to do
- 22:01something. It might seem like I did it for
- 22:02one reason, but I thought about 10 other reasons why to do it
- 22:05this way and why not to do it another way.
- 22:07And that might not be captured with screen recording.
- 22:10So I don't know, I feel like I would be massively disappointed
- 22:12by a current implementation of that.
- 22:15We might explore it at some point, but maybe that would be
- 22:17better suited for like browser automation tools that actually
- 22:20just click around for you in like internal portals and stuff.
- 22:23I'm sure that we have a Chrome extension that does that, that
- 22:26records your actions and replays it.
- 22:27It's a feature I don't really promote because it's a little
- 22:30hard to use, but I'm I'm sure that sort of thing is coming.
- 22:32I like that. Something to ideate for you, to
- 22:34watch you and then give you ideas for what you can automate.
- 22:37There's there's probably something there in terms of like
- 22:39a valuable product. What does your growth look like
- 22:41today? Our investors always tell us
- 22:43never to give numbers, but we are doing well.
- 22:46Like we grew 50% two months ago, 30% last month.
- 22:50It kind of were hitting a bit of an inflection point.
- 22:52Yeah, I mean, you're, you're on a tidal wave, right?
- 22:54And you've got a beautiful, beautifully designed product and
- 22:57it just works. What about like outside of gum
- 23:00loop? Are there any AI opportunities
- 23:01or business ideas that you're too focused on to chase but you
- 23:05think other people should be chasing down?
- 23:07Whether it's just a simple wrapper or something more
- 23:09complex? That's a good question.
- 23:11All I think. About is gum loop, it really it
- 23:14kind of encompasses my entire life and all of my thinking time
- 23:18a product that other people should pursue.
- 23:20Basically one that huge value problem gum loop is that all of
- 23:25these companies have this kind of top down directive of use AI
- 23:28and be more efficient. If you don't, if you're not
- 23:30using AI, you're going to fall behind.
- 23:32So we are kind of the spoon that is feeding them AI, but there's
- 23:34also tons of people at these really large companies that are
- 23:38are kind of resistant and don't want to move faster and like
- 23:42it's kind of hindering the growth of the company because
- 23:44they're living in their old ways.
- 23:46I think this is kind of like a controversial topic, but like
- 23:50one engineer with AI can have can do the the work of 20.
- 23:54Like I see that every day with my Co founder.
- 23:56He's like the most exceptional builder I've ever seen.
- 23:57We can move as fast as a large company because you just get 6
- 24:00hours to focus. So video is cool, but you know
- 24:02what's better? Long form audio via podcast and
- 24:05my newsletter, TKO pod.com. Go there to subscribe for free
- 24:09to my newsletter. It's one e-mail a week, very
- 24:11tactical. And then go to my audio podcast,
- 24:133 episodes a week. Stuff like this.
- 24:15You're going to love it all free, no sleazy sales bitch.
- 24:17TKO pod.com. If I was running a business that
- 24:20was kind of transitioning into AI would focus on those people.
- 24:23And I know LLMS can actually understand code at scale and
- 24:27understand what people are working on and flag people who
- 24:30are those TEDx engineers. Because I, I think like great
- 24:32businesses will be built by really small teams of
- 24:34exceptional people. So being able to identify which
- 24:37at a like a huge company with 500 or 1000 engineers, flagging
- 24:41the ones with, with AI that are exceptional, paying them double,
- 24:45you know, like treating them well so you don't lose those,
- 24:47those people. I think that's a really cool
- 24:49business that an engineering managers kind of tasked to do,
- 24:53but AI could do it at scale. Yeah.
- 24:55So it would basically be someone coming into a big company and
- 25:00using LLMS to read the code of their engineers and using
- 25:03certain markers to say, hey, this guy, this is your guy.
- 25:06Yeah, right. Exactly this guy.
- 25:08I think it's a code review, but honestly, like there's a lot of
- 25:11start-ups that do code review. Like you raise a poll request to
- 25:14introduce a new feature and it's actually like making sure
- 25:16there's no bugs. But it has such a deep
- 25:18understanding of the context of your code base and what the work
- 25:21you just did and what you added. And it can aggregate that over
- 25:24time. Make like weekly reports about
- 25:25who is really moving the needle for your business, and then you
- 25:28should never lose that person because they'll get paid more at
- 25:30another company that recognizes that they're great.
- 25:33So it's kind of a retention tool, especially in this age of
- 25:36AI where people are going to be chasing money.
- 25:39Exactly. I think it could be positioned
- 25:41as like a a cost cutting tool because you can like focus on
- 25:44not the good performance but the bad ones and then let them go.
- 25:46That's a more negative way to position it.
- 25:48But honestly, I think it's about rewarding the great people and
- 25:51putting them in a place of leadership so that like people
- 25:54will follow suit. Yeah, have you ever seen just a
- 25:57completely random business or use case for gum loop that just
- 26:01kind of made you cock your head and say huh?
- 26:03A completely random use case for gum loop that made me cock my
- 26:06head. In the early days we had some
- 26:08crazy hobbyist use cases. Like one guy recorded 3 hour
- 26:12sessions of Dungeons and Dragons, if you've heard of
- 26:14that, and then would use our our like video to text nodes in
- 26:18order to transcribe it. And then he would make chapters
- 26:21of the stories with a bunch of LLM steps that were kind of like
- 26:24digesting the content into this storyline, and then it would
- 26:26make a Google doc and send it to his team.
- 26:29So he was watching the three hour video and then yeah.
- 26:31It was just like his text, so you'd get this big blurb of what
- 26:34was said and then he was processing that in a bunch of
- 26:37like really nuanced, thoughtful ways, injecting like what the
- 26:40plot of their storyline is and stuff.
- 26:42And then he was getting this really thoughtful Google doc, as
- 26:44if he spent like 5 hours making like a post game summary for his
- 26:48clan. I thought that was pretty cool,
- 26:50really creative. That is interesting.
- 26:53Yeah. I, I feel like the the use case
- 26:55for AI watching video is still very early, but I, I feel like
- 26:59there's a lot of opportunity there.
- 27:00Totally. I mean a lot of government
- 27:02workflows are like video to blog posts.
- 27:05When I talk about with the content moderation stuff like
- 27:07someone records a 2 hour customer interview, they want to
- 27:10make a case study. You can actually drop that in a
- 27:12workflow that's like free set up in order to focus on what is the
- 27:15the most valuable 5 topics that they discussed.
- 27:19How do we structure this like our old content?
- 27:21You can set up that workflow and then every time you have a
- 27:22customer interview you can get the same quality of output by
- 27:25just dragging in a video file and pressing run.
- 27:27Wow. I mean, you could just integrate
- 27:28it with Fathom, right? And like anytime you have any
- 27:32Fathom call, you could just publish that to your website for
- 27:35SEO purposes. Totally.
- 27:37OK. Do you want to show me some use
- 27:38cases or some some flows and how to build them?
- 27:41Sure. Let me straight my screen.
- 27:45This was one that I always show off.
- 27:47This was like the one that. Changed everything for us in the
- 27:50early days when I figured out there's a scalable way to get
- 27:52meetings with customers. So I was really good at getting
- 27:55eyeballs in the product and people would try it out.
- 27:57But they had that problem we were discussing like they
- 27:58couldn't figure out what they wanted to automate.
- 28:00It took like a bit of discovery in order to list like 5 ideas
- 28:03that are really going to help them out.
- 28:05So this workflow takes in their e-mail.
- 28:08So every time someone signs up, I won't click on this because it
- 28:10actually shares my credentials, but it is it gives you a code
- 28:13snippet. So you can trigger this via like
- 28:15an API. Basically you're triggering it
- 28:17from your own code base. So in our case, when you press
- 28:20create an account on gum loop, it is throwing your e-mail into
- 28:23this workflow and pressing run and then that e-mail flows
- 28:26through all of these additional steps.
- 28:27So. Just to could.
- 28:29You zoom in on the additional 1. Yeah, let me at a high level,
- 28:32I'll just move this aside just to show you what gum loop like
- 28:34how things are structured. So we have all these
- 28:36integrations like let's say you wanna read emails from your
- 28:38Gmail inbox. You would drag a Gmail reading
- 28:41node, pick the inbox, and then I have all these emails to pass to
- 28:44another step. Let's say I wanted to categorize
- 28:47those emails. I would drag in an AI
- 28:49categorizer step, Next drag this in and then I would specify the
- 28:53categories, pick the model, add a bunch of additional context
- 28:57and then I'm doing GPG 4 O e-mail categorization and in
- 29:00like 2 clicks basically. So you're kind of.
- 29:02Connecting data with these nuanced AI steps and then you're
- 29:05able to build tools that would normally take an engineer like a
- 29:07couple days or a week in order to get like off the ground and
- 29:10deployed and usable for everyone.
- 29:13So this workflow is doing a couple interesting things.
- 29:16So it's not this node is not one you could find in our menu.
- 29:19It is what we call a subflow. So it is a flow within a flow.
- 29:23So if I click open subflow, I'll see that that that node is
- 29:26actually an entirely separate workflow that's doing research.
- 29:29So you're able to build these like composite creations that
- 29:31get more and more complex. And the cool thing is if you
- 29:34build like the best customer research flow that does the most
- 29:36nuanced research about your industry and your type of
- 29:39customer, your Co worker can use that workflow in their in their
- 29:41workflow as well. Like they can use that as a sub.
- 29:44So they're kind of like reusable tools.
- 29:46OK. Can you share flows between like
- 29:49completely unconnected gum loop accounts?
- 29:52Yes, yeah, you can share all flows like exactly like you
- 29:55would at Google Docs. So you would change the
- 29:56shareability like the ability settings, then you can send
- 30:00those around. So people are are sharing tons
- 30:01of flows. Do you have people selling
- 30:03flows? Yes.
- 30:05So we have gum loop experts, people who do contract work
- 30:07building for people. They they're kind of like flow
- 30:09builders for hire. We are going to have a
- 30:11marketplace at one point, but I want to make sure it's safe
- 30:14because it's kind of a unique product to expose really
- 30:18sensitive credentials and shareable tools.
- 30:20So I don't want someone accidentally leaking.
- 30:22Their liability? Box, Yeah.
- 30:23Yeah, now how are gumloop experts finding jobs today if
- 30:27you don't have a marketplace? We have an expert marketplace,
- 30:30it's an air table base, so it's like 10 different.
- 30:32People in this little directory and whenever someone reaches out
- 30:35to us for an expert, we actually introduce them to the experts
- 30:37directory in like a we are like the facilitator to send them
- 30:40that link, but it'll be much more official in the near
- 30:42future. Would you say as of today, is
- 30:45there more demand for experts or more supply than you need?
- 30:49For experts, more demand, so they're each getting.
- 30:51More experts. Yeah, but The thing is each
- 30:54expert has different expectations of like how much
- 30:56they should be paid. And so sometimes they might get
- 30:584 leads but only one is qualified to them in a given
- 31:01day. So they have enough inbound, but
- 31:03it does take some work in order to filter out the noise.
- 31:07They've got to close them like you're just connecting dots.
- 31:09You're not helping. Yeah, you're not helping.
- 31:11What are they charging on average?
- 31:13I know it's probably all over the map, but.
- 31:14It's all over the map. It could be like 100 bucks, it
- 31:17could be $5000. It really, it really varies
- 31:20depending on the scale of the work and what the expectations
- 31:23are. Yeah, sometimes people want a
- 31:25ton of work. They want to like totally
- 31:26automate their business for 200 bucks and that's just not
- 31:29possible. Yeah.
- 31:31It's about finding the right customers.
- 31:33Yeah, I mean, conversely, someone could build a flow for
- 31:36100 bucks and then that could that could turn into a job
- 31:39making $1000 a month, right, Really oxidating everything.
- 31:43So that could be one kind of unique way that people listening
- 31:45might be able to make money is just by learning to use it
- 31:47themselves as a user and then becoming an expert.
- 31:50It's a pretty interesting sort of transition.
- 31:51We actually hired one of our experts internally because he
- 31:55was just so good. Like he was blowing everyone
- 31:57away and I was like, what if you could do this for our largest
- 31:59enterprise accounts? You know, like when I really a
- 32:02large customer has a basic. Need he can hop on a call.
- 32:05Blow their mind with what AI can do.
- 32:06And all he had to do was like learn this drag and drop tool
- 32:09and now he has a new career, which is pretty cool.
- 32:11So their e-mail gets run through this subflow, which takes the
- 32:14e-mail, scrapes their website, extracts the name of the
- 32:17company, summarizes what the company does.
- 32:19It enriches all this information like industry, revenue and
- 32:22country because we're using all these different data providers.
- 32:23But you can get like monthly web traffic like I mentioned, or
- 32:26like LinkedIn URL formats, all of it in this nice little
- 32:28paragraph. And then that paragraph goes to
- 32:30our Slack. So like 10 seconds after you
- 32:33sign up, we get this notification saying how big your
- 32:34company is, what your base, what you do.
- 32:36And then it also drafts an outbound e-mail and into my
- 32:38Gmail inbox. Like really basic, like, hey,
- 32:40I'm the founder. Your company looked awesome.
- 32:42I saw you just signed up today. I'd love to chat here with my
- 32:44calendar sort of thing, but light touches, a personalization
- 32:47and just like, automated into my inbox.
- 32:48Yeah. I could, and it only sends that
- 32:50if it's kind of worth your time. Yeah.
- 32:52So our, the one that's actually in production is more complex.
- 32:55It's in our like private workspace.
- 32:56It has a bunch more qualification steps.
- 32:58So it does like much more research.
- 33:00And then it has the AI categorization steps.
- 33:01And then if they're lower value, I mean, just like hobbyist
- 33:04users, we put them into like a learning cohort sort of e-mail
- 33:08chain. And if not, then we do this
- 33:10personalized outreach for me so I can hop on a column and meet
- 33:13them. Do you have like plug and play
- 33:14flows that people can just tap into like this more complex ones
- 33:18or are they kind of basic? We have a whole templates
- 33:20directory. Honestly, templates feel like a
- 33:22Band-Aid over the problem to me. If you need a templates because
- 33:25you couldn't go from idea to exactly what you wanted with the
- 33:27tool, they might help spark ideas.
- 33:29But you can definitely get started with a template.
- 33:32I wouldn't like dissuade you. Yeah, I I really am excited for
- 33:35the future where I can go from an idea to a workflow in 5
- 33:39minutes without ever having to scratch my head or assuming I
- 33:42was a new user. Now what about like sales
- 33:45prospecting? I think it's this workflow has a
- 33:48couple of those properties. So this one before 15 minutes
- 33:52before every meeting, it will look at the attendees, filter
- 33:56out all the ones that aren't me or internal people that come
- 33:58loop. It'll read my e-mail inbox.
- 34:00So this is kind of like a sales workflow.
- 34:01Like when I'm hopping on a demo call with a big customer, I want
- 34:04to know why the e-mail was set up or why the the meeting was
- 34:06set up. I want to know what the person's
- 34:08role is at the company, what they do, their background, any
- 34:11socials I can possibly scour so I can understand more about who
- 34:14they are. And then I want to understand
- 34:15what the company does. So like this research company
- 34:17step, this work was a little complex because I actually just
- 34:21use it all the time. But this flow here will take
- 34:26their e-mail, search their inbox for all the reasonings for why
- 34:29we might be meeting. And then it has two different
- 34:31outputs, one being research about the person I'm meeting and
- 34:34one being the research from my e-mail inbox.
- 34:37So I get this like big digest, just like you would go to your
- 34:39Gmail, look at the 50 most recent emails with you and mark
- 34:42and figure out like, oh, we actually set this meeting up two
- 34:44months ago. It's it's about talking about
- 34:46like growth or something. So you have an LLM do that sort
- 34:49of digesting and then here we do company research.
- 34:52So this is another complex one, but this is where the crux of it
- 34:55is just this AI web research node.
- 34:58So given a company's URL, identify the company's name,
- 35:00their current stage, whether they're pre launch seed growth
- 35:02or non VC backed, find their LinkedIn and their Twitter URL.
- 35:05And this sort of information is what I normally just check on my
- 35:08phone before the meeting to understand like who am I
- 35:09actually talking to today? There's a bunch of like
- 35:11conditional pads here because I sometimes people don't have
- 35:15social, there's edge cases that I built into the workflow.
- 35:18And then all of that gets synthesized into one big prompt
- 35:21where I say essentially like, write a text, take all of this
- 35:24information and give me the most important things that I need to
- 35:27know before the call, and then send me my text notification
- 35:30here to my phone. Well, seeing these ideas just
- 35:34give me so many more ideas. And it's gonna be cool when you
- 35:36can, when you have this idea, you can go to your keyboard and
- 35:40get the actual tool built in like 2 minutes.
- 35:43That's gonna be super exciting. This is another random one.
- 35:45I made this with a pretty large HR software company.
- 35:49I won't say which one, but essentially what it's doing is
- 35:53there's these things called gumnip interfaces.
- 35:56I'm not sure if you've seen these, but this is like a UI
- 35:57layer on top of your flow. So you can abstract away all the
- 36:00crazy workflow stuff and just have an interface.
- 36:03You would put a competitor domain and then it uses Semrush
- 36:09gets the 100 keywords that that competitor ranks for.
- 36:11So we built this because this this SEO expert was having to
- 36:14look at competitors, look at all the keywords they rank for and
- 36:17then figure out which ones aren't name like brand related.
- 36:20So if it's like a Samsung and the 50 top keywords they're
- 36:25ranking for out of the top 500 are like Samsung or misspellings
- 36:28of the word Samsung, they like erase those from the list and
- 36:31focus on the other ones. But that sort of reasoning like
- 36:33is this a brand keyword? Is it not at scale is so time
- 36:36consuming. So they just built a workflow
- 36:38and I think this actually took like 4 minutes to put together
- 36:42and it was taking them hours every day because they're doing
- 36:44this for so many different brands, so many different
- 36:46keywords, and then it dumps all of the non brand ones into a
- 36:49Google Sheet and then they start doing all this other automation
- 36:51on those keywords. Man, do you have anyone using
- 36:55gumloop to build like programmatic SEO sites?
- 36:59Yes, definitely. This was 1 workflow we're headed
- 37:02over here. So I mean like posting and
- 37:05making structured content is great on the platform.
- 37:08So this takes a YouTube URL. This is like an hour and a half
- 37:11long Stanford lecture. I think about what LLMS are
- 37:14running the workflow here. The blog is live right here
- 37:17already. And this is like a really Gen.
- 37:21Z sort of like post but trying to make a boring topic
- 37:24interesting. All of that is engineered into
- 37:26this workflow here, which is just like a three-step O1
- 37:30workflow, taking the YouTube content, making really good blog
- 37:33sort of output from it. And then the cool thing is you
- 37:36can take this whole workflow and run it at scale by looping it
- 37:39over a Google Sheet of 1000 links if you want.
- 37:42So you can make your like the same reasoning you would apply
- 37:44to make the perfect programmatic SEO page for like one topic.
- 37:47You could feed in 800 topics and apply the same reasoning just
- 37:50overtime. Go get a coffee and come back
- 37:52and all your contents then. So for instance, we have a
- 37:55programmatic site for pet cremation leads and we have 5500
- 38:00landing pages which represents the top 1100 cities in the US
- 38:03and Canada by population. And then we have like 3
- 38:06keywords, pet cremation, horse cremation, pet euthanasia, and
- 38:10it's all in web flow. You're saying that we could put
- 38:12like all the keywords in a Google sheet, all the city names
- 38:16in a Google sheet, a prompt, maybe the same sheet, different
- 38:19columns, a prompt that says like, hey, give us a fun fact
- 38:23about this city and then tell us the population of this city and
- 38:26something unique. And it can just generate all
- 38:29these different landing pages for so every so no two, No 2
- 38:32landing pages are the same A. 100%, yeah.
- 38:36Also the founder of Webflow is, is an advisor to Gumloop because
- 38:38they're big users. The guy who was doing education
- 38:41at Webflow, most recently, Aaron, he joined our team.
- 38:44So he does education for Gumloop now because he loves Gumloop so
- 38:46much. That's amazing.
- 38:48But you can totally do that. That's a slam dunk use case.
- 38:52I mean people, people charge a lot of money for a programmatic
- 38:56SEO as a service and it could just be built on top of this
- 38:59without much technical knowledge.
- 39:01Exactly, you just play with some Lego blocks and get the value.
- 39:05What about the the meta ads one? This one is really cool.
- 39:09I mean, you put in a link to your competitors Facebook page.
- 39:12It gets all of the ads that running off Instagram, WhatsApp,
- 39:14Facebook, like the whole meta stack.
- 39:16It runs all of those through Gemini to watch the ad, look at
- 39:20the copy, tell me what they're advertising.
- 39:21Like what is this strategy for this specific ad?
- 39:23And then for let's say there's fifty of them.
- 39:25It stitches all of that together into one massive prompt, all the
- 39:27descriptions and runs it through A1, which is just like a
- 39:29thoughtful large model. And it makes a report saying
- 39:33what this editor is focusing on from an ad's perspective right
- 39:36now. So you get this thoughtful
- 39:38report that's formatted with logos and stuff and links
- 39:40embedded like the whole research report in your inbox.
- 39:42And you can schedule it to run every Friday at 9:00 AM.
- 39:45So you get like the weeks update on your competitors ads right?
- 39:48Oh my gosh. I mean, every single one of
- 39:50these flows is a business. Like that's how I see the world,
- 39:53right? Like you see the world as gum
- 39:54loop. And you should.
- 39:55That's what you're building. But like I see this is like I
- 39:58can, I can start a Facebook ads reverse engineering as a
- 40:01service, right? Pay me $300 a month, I'll send
- 40:04you an e-mail every Friday with your top three competitors.
- 40:07It's going to tell you are they doing 34% display ads, 13%
- 40:10stories? You know Facebook, Instagram, is
- 40:13it Instagram video post carousel fully automated and.
- 40:17As far as your customers know, you're spending four hours a
- 40:19week doing this for them by hand.
- 40:21It's awesome. And then you can also do it for
- 40:23like 10 different competitors and string that into one larger
- 40:26report so you can do the comparative analysis against
- 40:28like 6 at the same time or 10 at the same time.
- 40:30What I would like to use this for is I have a paid, I have a
- 40:33newsletter, a free newsletter. But sometimes I use Meta ads to
- 40:36acquire subscribers. And I would love it to run a
- 40:38report from the Facebook ad library every Friday and say,
- 40:42show me any Facebook or Meta ads that have been active for six
- 40:46months, which tells, I mean, that they're high performing ads
- 40:49that have the word newsletter or podcast in them.
- 40:51And then every week I can just kind of see what other people
- 40:53are doing, like what ads have been around for six months, AKA
- 40:56what ads are performing really well.
- 40:58Or you can even have it cross reference the UK version of the
- 41:01Meta ads library because in the UK they have to disclose like
- 41:05how much money people are spending on their ads.
- 41:07This is like the the reason government exists.
- 41:09So like, I would never have known that an ad that is up for
- 41:12six months is valuable, or that the UK has to give you the ad
- 41:15spend that people have on it. But if you have that knowledge
- 41:17and you know how to use tools, you can build software that is
- 41:20immensely valuable without needing to be an engineer.
- 41:23It's just a matter of like applying niche knowledge to
- 41:25tooling. I feel like there needs to be a
- 41:27whole community built around just people, people starting
- 41:30businesses on gumboop, not just using it right, like that's
- 41:33sure, but like starting businesses, reselling services,
- 41:37white labeling gumloop in in unique or even I think white
- 41:42labeling gum loop in simple ways is even more unique, you know?
- 41:46Yeah, yeah. I mean if you use our web hook
- 41:48you can build your own web flow page for example.
- 41:51With a form. Where you put in like the
- 41:53competitors ad Facebook ad link and then the e-mail will be sent
- 41:57and they don't even know gum loops involved.
- 41:58It's just powering the product that is customer facing for you.
- 42:01So white labeling is is really easy in that regard.
- 42:04Oh, that's beautiful. Well, this has been awesome.
- 42:07Is there anything else you want to leave us with or show us?
- 42:10There's a bunch of features I didn't show off, like the
- 42:12ability to make custom integrations with LLMS. That's.
- 42:15That might have a question for you too, but yeah, please.
- 42:19I won't go too deep into it. It's going to look different in
- 42:21a week from now, so there's no point demoing it.
- 42:22But essentially like if we don't have an integration that you
- 42:24really want, that's going to change.
- 42:27We are. We're scaling the integrations
- 42:28first and foremost with this new MCP paradigm that I haven't
- 42:31touched on. But also you can make your own
- 42:33custom integration that is just a prompt away.
- 42:35Basically you ask for what the API you wanna hit is, or like
- 42:39the third party we're not integrated with.
- 42:41And then it will write the code for you, deploy it for your team
- 42:42and then you have this like totally custom little tool.
- 42:45Kind of like a mini replit inside of gumloop.
- 42:47Exactly. Yeah, code Gen., but simplified
- 42:49into workflows. That's powerful.
- 42:52So the question I had is when you were describing the the meta
- 42:56ads analyzer, you said you send it through Gemini first, then
- 42:58you finish with A1. Like what do you suggest people
- 43:01use? What models do you suggest
- 43:03people use for different tasks? When do you use Gemini?
- 43:06When do you use Clod? Etcetera.
- 43:09Good question. It really is an artistic choice
- 43:12I think. Like it's a preference.
- 43:13Like some people like Pepsi, some people like Diet Coke, some
- 43:15people like Clod, some people. Like open the eye.
- 43:17I always go for the most powerful model when I need the
- 43:20most reasoning. So for like the analysis for
- 43:23each ad, I might use a G PT4-O because it's it's smart.
- 43:26And then when I want to do the analysis over 50 ads from the
- 43:29GPT 4 analysis, like stitched together, that's more of a long
- 43:32term or long form reasoning task that needs an O one or would
- 43:36perform better with an O1. And then if like let's say
- 43:39Instacart is labeling 80,000 lines of produce, I would say
- 43:42don't use A1 for each labeling task, use like GPT. 4-O mini or
- 43:47Gemini 2 point O or Gemini 2.5 now.
- 43:50Whatever they call it, Gemini Flash, use like a cheaper model
- 43:53that is faster cuz sometimes the task is so easy.
- 43:56It's like there's a produce label, there's the name of the
- 44:00item and there's an image. And I'm saying like, is this
- 44:02beer supposed to be labeled produce?
- 44:04Like that is an obvious task. So I use a cheap fast model for
- 44:07that, and then it costs almost no money to get reasoning at
- 44:09scale. What about just for web
- 44:11research? Like for your inbound leads and
- 44:13then for writing emails. I am not really cost sensitive
- 44:17like. All of these work flows are so
- 44:18important to me that I would rather have better reasoning
- 44:21than just save like. Half a cent by using a worst
- 44:23model, so I'll normally just stick with GPD 4, O or Claude.
- 44:273.7 is exceptional. OK for either one of those.
- 44:32Yeah, I normally like just switch between them.
- 44:34I think Claude is a little bit better at the creative writing
- 44:37style stuff, so I use it for that.
- 44:39But some people would probably disagree with me.
- 44:42Awesome. Well, this is perfect, Max.
- 44:44Thank you for your time. Obviously, we can find you at
- 44:46gumloop.com. If people want to reach out to
- 44:48you, where can they find you? I'm on Twitter a lot, it's my
- 44:51full name so just Max Broder urbis maybe it's.
- 44:54We'll link to it. Sweet.
- 44:55I'm pretty active there. Same thing with LinkedIn.
- 44:58Full name, yeah. OK.
- 45:00Well, thank you for your time. Sweet, great to meet you all.
- 45:02Right, what do you think? Please share it with a friend
- 45:04and we'll see you next time on the Kerner office.