Building AI Agents for Everyone

Author: YC Root Access · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Building AI Agents for Everyone · Channel: YC Root Access · 33:56 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Task and workflow automation, AI governance and regulation, Markets and economy

Summary

Gumloop’s founders explain how the company evolved from a visual workflow automation tool into an enterprise platform for building, sharing, and governing AI agents. They describe how customer feedback, an early enterprise champion, and improving models shaped the product and its focus on security, integrations, and usability. The conversation covers how companies adopt agents, how Gumloop uses them internally, and why letting teams choose models while retaining control over their agents matters. The founders also discuss pricing, fundraising, hiring, and the role of San Francisco in building momentum.

Key points

Gumloop began as a visual workflow automation platform because early models were not capable enough to run agents reliably; as models improved, the product shifted toward agents that can automate tasks end to end.
The platform connects to company tools and can bring agents into places such as Slack and Microsoft Teams, where teams can use and share them.
Common use cases include sales agents that handle research, CRM updates, call analysis, and follow-ups, as well as customer success and internal data agents.
Enterprise adoption requires features such as role-based access control, SAML, audit logging, deployment options, and fine-grained data permissions.
The founders learned about customer needs through extensive calls, workshops, templates, and hands-on onboarding; they found that enterprise use cases often delivered greater value than early individual use cases.
A power user at Instacart helped introduce Gumloop internally and later joined the company, helping the founders understand enterprise sales and procurement.
Gumloop aims to let the people who understand a task build the agent for it, while IT teams provide the data access, permissions, and governance needed to make that possible.
The company uses agents internally for tasks such as answering company-wide questions, supporting sales and customer success, and automating educational programs.
Gumloop offers model choice and transparent usage-based pricing, positioning itself as infrastructure that customers can deploy in their own cloud and use without being locked to one model provider.
The founders say agents can help employees do more rather than simply reduce headcount, and identify awareness, hiring, and the complexity of reliable long-running automation as ongoing challenges.

Chapters

Start	Chapter	Summary
00:00	Introducing Gumloop and its evolution	The founders introduce Gumloop as an agent platform for companies and explain how it grew from a workflow builder as AI models improved.
02:04	Agent use cases and enterprise customers	They describe sales agents, integrations with existing software, early users, and the shift toward higher-value enterprise needs.
05:09	Finding product-market fit and winning enterprise adoption	The founders recount learning from customer calls, landing Instacart through a power user, and building the features large companies require.
09:45	Templates, onboarding, and the founding story	They discuss helping customers find use cases, then tell how they met in Vancouver and came together to apply to YC.
13:23	Making companies AI-native	The conversation turns to adoption, collaboration, and how Gumloop uses shared agents to bring AI into employees’ existing workflows.
16:29	Recognizing demand and growing through customer rollouts	The founders recall users investing hours in the product and describe the Shopify rollout and varied patterns of company-wide use.
19:02	Pricing, fundraising, and company building	They explain the move from seat-based pricing to transparent usage-based pricing, their fundraising approach, and decisions about location and team size.
24:39	Who should build agents and what holds adoption back	The founders argue that people closest to a task should automate it, with IT enabling access and governance, and discuss agents as a way to expand employees’ output.
27:43	Building reliable agents and preserving customer choice	They outline technical challenges such as long-running tasks and browser automation, and explain the value of model flexibility and customer ownership.
30:20	Model selection and enterprise sales	They discuss matching model capability and cost to each use case and how adoption can move from an individual champion to IT leadership.
32:24	San Francisco, motivation, and closing reflections	The founders reflect on the momentum they gained in San Francisco and the motivation to work quickly and avoid missing an opportunity.

Quotes

So we were a workflow automation platform. — 01:02

I took 1100 customer calls the first year. — 05:40

bringing agents to where they’re already doing work has been the biggest unlock for us. — 14:25

the people who understand the tasks should be the ones automating them. — 25:41

owning your agents is quite important. — 29:49

Concepts and entities

Gumloop, Max, Rahul, YC, Shopify, Instacart, workflow automation, Agent, MCP, role-based access control, Heimdoll, Salesforce

Full transcript (original language)

00:00 [music] Today I’m here with Max and Rahul from Gum Loop. Um, thanks for coming back to IC. >> Thank you for having us. >> I’m really excited for this conversation. You guys are one of the most successful yet kind of unknown companies uh from Win24. Tell us what Gum Gumloop is. >> Gumloop’s the agent builder that a lot of companies like Shopify and Gusto and Instacart uh use so that everyone can build agents and share them across the business while it can manage and control that data. So kind of like the holy grail of agents like this is what

00:31 everyone is talking about what they want to actually actually build like this is probably one of the more common applications I guess we get from Twice today but you actually built it and you’re deployed in like many of the biggest companies. >> Yeah it’s definitely a crowded space because it it feels like obvious that someone has to solve this problem. There’s definitely a bunch of players that are doing a good job but we’re just trying to >> be one of the leaders be the Yeah. And you applied to IC for uh winter 24 uh you called agent hub at the time >> and then I switched to gum loop. >> Yeah. >> Um what was this product when you

01:02 applied? >> Yeah, it is ironic that we called ourselves agent hub back then because now we are an agent hub uh but there was a bunch of everyone thought we were a real estate agent company because agents weren’t common uh right >> back in the like back then >> but we were a workflow automation platform. So same mission of like automating tasks but the models weren’t good enough back then. So we just made a visual builder and people could define their their workflow step by step and then we added agents and that was kind of like the hockey stick moment where like anyone could build it on the platform and usage kind of skyrocketed but um same mission since the beginning.

01:33 >> How does gumlook work like how would you actually use it? >> Yeah. So basically you can go onto the platform and then create an agent um and add >> how do you create it? >> You create an agent in the onboarding but you can also just like click the create agent button and then you’re presented with like um the kind of standard like chat interface. Then you have like a set of connectors that you can kind of like pick and choose all the apps that you work with. Um you can start talking to the agent right away to get work done or you can add the agent to your Slack workspace, Microsoft Teams, uh wherever you work, we kind of bring the agents to you. >> What’s the most like what’s some good

02:04 examples of common agents that people make once they have the your UI? >> They tend to be like um team specific. So the we the sales team at pretty much all of our customers for example will have like a sales agent that’s connected to like their call recording software, their CRM, a bunch of enrichment tools, web search tools, and it has a bunch of skills for like how they like to do business and how they want to manage like opportunities and analyze calls and uh it it becomes like a co-orker for the sales team that the sales team uses and relies on. So they’re pinging it in Slack every day. They’re CCing it on

02:34 email threads. They’re having it set up next steps, clean, dduplicate accounts, like all of the tedious work that a salesperson would be doing. They built an agent to do that for them and it’s it’s like their little co-pilot. >> Does it replace the software tools that they currently using or is it kind of interacting with them? Like how does it work with say Salesforce or something like that? >> Yeah, we like to be the connector between all of your tools. Uh so sometimes it actually does replace like vertical SAS you might have procured for a lot of our customers. They’re like canceling contracts to just use an agent instead. But often times we’re like the the gum we like to say between your data different data connectors. So you’re

03:06 using all the tools that you love but you’re bringing them into one place and giving it to like this this like alien like little co-pilot that can just do stuff for you in the background. >> A lot of companies are developing internal tooling. Their own engineering teams are building agents and we kind of fit right into that ecosystem as well. Like we provide a pretty robust SDK and API. So pretty much anything you can do through the UI in Gum Loop you can do via API or SDK and we have like hundreds of integrations exposed via MCP servers. And so a lot of our customers are building internal tooling and infrastructure using Gum Loop as a base

03:37 layer for their AI. >> And it seemed like you were one of those companies where basically you had this idea before models were good and once model got good this just like this product got like infinitely better. >> Is that what happened? Yeah, I think that was definitely one of the strategies at first which was like the models were not good enough to make agents work and so we kind of met the models where they were and that’s why we started with this workflow builder and kind of like minimized the use of AI and just had it in the key area that it needed to be used for like a giant workflow. Everything else was standard infrastructure that made things actually

04:08 reliable and economical at a point where the models weren’t good enough. But now models are good enough to automate tasks completely end to end. So you don’t really need this workflow manually built out. the agent can just do it by itself. >> Who were the first users? What were they building? >> There was like an initial wave of like PLG growth. So like individuals, indie hackers, people like just having really tedious tasks before AI could really automate anything. >> Solo like solo. >> Yeah, exactly. Like I think during our YC interview, the people the use case we quoted was someone transcribing their Dungeons and Dragons session. Uh that

04:38 was like innovative at the time. The fact that they could talk for two hours and then have a transcript and talk like send it around to their the attendees. They were using a gum loop workflow for that. But then little use cases like that. And then all of a sudden we had business users like I remember this Australian health tech company was using us to automate a lot of their research just by like pulling from the web and like regulating that data and putting it in certain places. And then we realized like the enterprise customers are willing to pay a lot more. Their problems are a lot more serious. Um and like the value we provide them is just

05:09 so much deeper. >> How did that realization come about? slowly over time it I think we always wanted to be the PLG tool that like your your mom would use and like that your coworker would be using but we just saw like when you were giving the pitch to different people like the people who have a thousand employees and they need like a layer to help them all automate and they deeply need security and observability and access control and all of like the boring stuff behind automation like they start their eyebrows raised higher than like the uh >> but I’m assuming you didn’t get to that pitch right away there was some some step function to get there what that

05:40 what was that like >> I think I took 1100 customer calls the first year. Uh I was like top.1% of cal.com users >> and uh it was iterative like every call you’d kind of hone in on what is getting someone excited what makes them nod their head and it all trends towards like what we’ve built now which is the enterprise platform >> and when the product wasn’t good back then we really had to be super involved with customers and help them to get something into a working state right so we were like very familiar with their use cases and what exactly what they were they were doing and a lot of these solopreneurs or like these consulting

06:11 type people it was just really hard to make things work for for them and even when we did make things work for them, it wasn’t generating that much value. But when we saw, you know, like this Australian health tech firm, when we made their use case work and we s saw how much money it was like making them and saving them, we were like, okay, this is like obviously the natural direction of things, but I don’t think it was super clear at the beginning. >> Instacart was your first real enterprise. How did that happen? We had one power user there who was like obsessed with gum loop and uh started bringing us he brought us into a slack

06:42 channel and then he was kind of like selling the tool for us internally like starting all these threads because he was the one teaching everyone at Instacart how to like use AI tools and how to automate tasks. I think he was he was using every tool in the market to like show people how to be more efficient because he saw the light of like what AI was capable of before most people. Um, and then we ended up hiring him as one of our first employees and he helped us like understand what enterprise sales means and how to talk to a procurement team and how to price and >> what a champion was. >> Exactly. What how to identify a champion and and then that led to our next logo and next logo and then um people at

07:13 these big companies go to other companies and if they love your tool enough they’ll bring it with them. So we just had this like natural growth uh that happened in the early days. >> Were you the best product? Were you like the just first product? like how how did you grab like have a foot foothold in in big tech companies that other companies don’t have? How did that happen? >> I think for the workflow automation builder, we genuinely were the best product. We’re by far the easiest product to use and we our product allowed you to do things that you couldn’t do on other platforms and a um some of the people Aaron who’s our head of community and education uh he’s taught a bunch of these no code tools

07:44 and the one thing that he always said to us was um the reason that Gum Loop is so capable is cuz you guys did not know what any other product in your space did or looked like and we kind of built things from like first principles and so the way our thing worked was just a little different than anything else and it made it so much more powerful and we realized quickly why other tools don’t work that way because it’s actually really hard. The problems that we were solving were really hard. >> Yeah. Tell us about these things. So, so, so when you sell to a big enterprise, there are things that they

08:14 require that small companies don’t. Like for someone who’s like watching this and want to sell to enterprise, what are those things specifically? >> Yeah, I think stuff like role-based access control, SKIM, SAML, um, [clears throat] audit logging, all that type of stuff is like um, >> why do big companies ask for that? There’s just so much at stake when you’re a large company. Like you’re a publicly traded establishment that has like sensitive data and all these requirements and like restrictions with what you can do. Uh so you need to make sure that if you’re going to give 10,000 people at your company a tool, they’re only able to do the things you want them

08:44 to be able to do. And certain users, it gets infinitely detailed like how they’re like this these 500 users need to be able to do these 10 things. These users are actually an outsource team that shouldn’t have access to any data. We need them to have restrictions on their accounts. We need to use our own API keys. We need to host it in our own cloud. We need to have audit logging for every tool. We need to have APIs to get that into our data lake. Like it never ends. Uh but that’s like the value we have to create before the end user can have that wonderful moment of like, oh, I connected data and it just worked. Like there’s so much depth before they can use it. >> Yeah, it can be frustrating sometimes

09:15 because they want to disable so much functionality that we worked so hard to build. But uh yeah, I guess that’s what it takes. And then when they warm up to us, we just enable more and more things. >> How did you figure out what all these things are and how to h what they need? because like you can’t show up and be like, “Oh, we don’t have these things. Let’s go build them.” I’m assuming yeah, you have to figure out how long >> there’s a it’s just conversations with customers, like hundreds of conversations and um you get a sense after building for a while of like the direction like what feels directionally correct when they’re asking for a feature and what you should just ignore. Most of the time like everything that

09:45 our big companies are asking for is something other big companies would want at some point. So, we build it. Uh, and I think like the level of care you have and the speed at which you respond to their like the amount you feel like a partner to them and the speed that you ship what they want it really makes a difference. >> One of the things I’ve noticed with companies and I remember we had these conversations early on is if you have a tool that you can do anything people don’t really know where to start. >> Yeah. >> How have you solved that today and what are the most common thing that people use your tool to do? >> Yeah, that’s a good question. So, I remember when we met you for the first time, when you saw the product, the

10:16 first thing that you said was like, “You guys need templates, like pre-built use cases to show people what to do.” And and then we came back the next week with like three different templates. And then you were basically like, “Guys, that’s not what I meant. You need like a hundred of these things.” And so, we built this like really large template gallery. We kind of do similar stuff like that today in which we have like use case pages with like pre-built agents and that kind of shows gives people an idea. Case studies help. We do workshops with companies and we’re like tell us if you could wave a magic wand like what would you automate? We get like 20 hands raised and then we’ll help them all automate their own things and

10:47 then that sparks more ideas. >> You handhold companies through the onboarding process to kind of build out the initial >> sometimes they like come to us with like 10 ideas and they’re like I already did five. These are five more. I need like these features shipped before I can do that. And we’re like okay consider it done. Um but sometimes it requires like fully inspiring a team. >> Let’s go back to like earlier prior to doing YC how do you guys meet and decide to start coming together? There was an AI conference going on in Vancouver and I was presenting what I was working on at the time which was like a personalized AI tutor for students and Max was presenting an early version of Gum Loop and uh what he was building

11:18 sounded really cool. It sounded really promising. I thought it had real potential to work and so I basically like asked him like if he needs a co-founder. Well actually he said there’s a couple caveats to this story. There’s one the conference is a very generous term. It was like 30 people in a room uh just like talking about AI before like right when the GPT API came out. Basically, my co-founder had quit that morning. The guy I was working with, who I met on YC, co-founder matching. >> And he was like, “This idea won’t work. I need someone more senior.” Like this whole laundry list of reasons. I believe the no, but not the why. I learned that after working with YC. But I went to

11:49 this meetup to take my mind off things cuz I was so like distraught and I’m like, I’m screwed. Basically, there’s a bunch of other things that happened in the background that made me more screwed at the time. >> Rahul was demoing with his co-founder. Uh, and that person ended up joining as our founding engineer like a year later. But um the whole video of us interacting for the first time is captured on my Twitter. I was like, “My co-founder quit this morning. I’m looking for another one.” And then the next person to speak is Rahul asking about GPT functions. Yeah. Which had just come out that day and he’s like, “Have you considered those?” And I’m like, “No, I don’t even know what that is.” >> But it was very serendipitous. And then we got drinks the next day and he came

12:20 with like his laptop that was duct taped together >> [laughter] >> uh with like a Google doc of like 50 questions because the product was open source at the time. So he read all the code before AI could just answer questions about like a codebase. and he had like very detailed questions about the framework that I had built. So I was like this guy’s pretty serious. And then we worked in the co-working space together for like a month. >> And you were both in Vancouver. >> Yeah. In Canada. >> And then from there to applying to YC. >> Yeah. I think we always wanted to apply YC. Max knew way more about it. He had applied to YC three times in the past. I had never applied to YC, but we knew

12:51 what YC was. And so >> I’d applied with very different ideas previously. And I’m actually thankful in hindsight I didn’t get in because it would have put me down the wrong path like working on ideas that weren’t the right fit. >> Uh but yeah, this was my fourth time applying. And um we worked in a co-working space like for 12 hours a day every day for a month and a half I think until we applied to YC. >> And at that point we had like people who love the product a very small number of people who love the product but like they got value out of it >> and we got lucky and interviewed >> and how I see we we’ve released something called uh quota master

13:23 recently is open source. We’ve built our internal company brand at YC. >> I am like this is the biggest software tool we ever launched at YC internally. Like I use it all the time >> and I’m assuming that your customers are kind of like this it might be the biggest thing that they ever seen internally in terms of productivity. However, it takes just like hard to get people to start using things. I’m I imagine that you also built up this knowledge and how to basically make a company AI native that others don’t have. And this is sort of like at the top of the list of every management team right now is like how do I get my

13:53 company to be become AI and talk about that like like >> this is like the holy grail again in like the trans transition or trans transformation of AI. >> I think um that was the first chapter like we definitely built up a lot of expertise around making a company AI native and use AI for as much as you can and then there’s the second chapter that comes immediately after that where you’re like okay they’re using it for too much and we need to like actually govern and moderate and make sure it’s being used in a productive way. So we’ve been we’ve gone through I think we’ve built expertise in both but tricks for making your company AI native bringing agents to where they’re already doing work has been the biggest unlock for us. So like if you have to go to yet another

14:25 platform and use the AI system there there’s friction there. But if you bring it to like the the platform they’re already spending eight hours their day in you they’ll natively adopt the tool. I think bringing it to a place where you can see other people use agents and see how they use agents. I think like that’s one of the things that Gum Loop does really well compared to a lot of other agent platforms is like this aspect of collaboration aspect of like sharing agents, multiple people talking to the same agent, that kind of multiplayer portion of it. And being able to see how your co-orker who does literally the exact same job as you interacts with an agent in a public Slack channel

14:56 naturally lends yourself to using agents in that way basically. >> Got it. Can you talk about how you use Gumloop internally at Gum Loop? >> Yeah, I think everyone on every team uses it for everything. The first like breakthrough agent that changed the way we work was this one we call Heimdoll which is connected to every single data source at the company. >> Um and it has a Slack channel that was just ask Heimdoll originally but now it’s like in every single channel across the entire company and you can ask it anything like get me all the users that signed up yesterday morning that are in this time zone that have this role and

15:27 uh aren’t in this certain ca it’ll like work for 20 minutes and get you that answer. we don’t need like a data analy analysis or analyst >> analyst >> um we just have like Heimdoll answering every customer related question and then that spawned a bunch of ideas like the sales team went off in their direction and they have sales loop which automates everything in the sales process so they can just focus on talking to customers our customer success team has uh success loop which like the health score of every single customer is like diligently tracked they get notifications if anything’s ever going wrong if any usage

15:58 is skyrocketing or dipping like everything that they were spending their time monitoring is handled for them. >> Yeah. >> Our education lead, we have cohorts every two weeks where like he teaches hundreds of people how to build agents. The entire process is automated from like people signing up, the follow-ups, the uh even the editing of the videos when they demo on Fridays and show their agents fully automated so that the clips get sent to those customers and they can share it around their companies. >> Throughout this process, was there a moment when you felt we have PMF, this is work working like like was there something like very visceral that you remember or was this more gradual? I

16:29 think there’s been a bunch of moments like that, but I remember the first one uh felt like our first conscious thought about whether this is something that people actually want was when they were using our initial workflow builder early early days and it was awful but they were using it anyways, >> right? >> And we’re like we’re actually I’ve built a lot of bad software before that no one used and for some reason people were like spending five hours of their afternoon like building a workflow on our platform and we’re like oh this is actually so painful for people that they’re going to they’re willing to put up with how bad this is. And um I think that feeling of like wow people are

16:59 really investing their time has hit us every couple months. But I don’t know if you have anything else. >> Yeah, I think like when we realized that certain users were spending like eight hours a day, 10 hours a day using Gum Loop building stuff like automating their entire jobs or like the job of their entire team. That was I think big. It was when we realized like yeah our users are spending way way more hours than we are in gum loop you know building stuff and like some of them know how to use the platform better than you know we do in some ways. I think that for me one of the big moments was

17:29 like the first day we launched that Shopify because it was kind of like um we basically just like went out to like 4,000 people on the same day. >> I think it was 8,000 people. >> 8,000 people on the same day. >> They just got access in one day. just seeing people kind of like pick up and like use the tool and like play around. It was like it was a super cool moment and people were just like figuring things out and like we didn’t have to I think before then it was a lot of like sitting down with customers and like building stuff out out for them and like a lot of handholding but then yeah I

17:59 don’t know people just started building stuff on their own which was cool. >> Yeah. >> How did you get Shopify? I I I remember Toby writing the AI native sort of like sort of like his manifesto almost. Uh it sounds like could not be a better customer than them. How how did you get >> It was actually that was the one of the word of mouth customers where like someone from Instacart had influence at Shopify and just put it on their radar and then they evaluated the tool. I think they um like rolled out a bunch of tools the same week like different automation tools and they just saw what’s stuck. >> Um and then they ended up investing in Gum Loop after because they they really liked >> the roll out. Um and they’ve been a

18:30 great partner since. >> Got it. What percent of employees of these companies are using Gum Loop on like a daily weekly basis? >> It varies a ton by company. Sometimes it’s like extremely concentrated usage with one team, like everyone using agents to automate this one function. Sometimes it’s like wall-to-wall deployments with like light usage on some teams and deep usage on others. Sometimes people are using Gum Loop, but they don’t even realize it. Like they’re getting all these reports and stuff so that they can do their job better or like the prep for a customer call or something, but it’s it’s all gum loop in the background. So there’s like it’s uh completely different patterns at every company. But for the most part, it’s uh

19:02 everyone has access. So like we never charge per seat. I never want someone to like second guess whether they should be using this tool or not. It’s just usage based. Um so whoever wants to start automating can just turn to Gum Loop and figure it out. >> That’s not how you get started though. When did you switch to usage based pricing compared? Initially it was seat based. >> Yeah, we had a limit on the team plan for seats. >> How did pricing develop over the years? >> Yeah, I think when we first started we had uh a $20 a month plan and a $40 a month plan and that’s what we went into YC with. And uh I guess we picked those numbers cuz like chat GBT had something similar and that was kind of like all we

19:33 knew is like products that what we like to use were in that price range. But then we quickly realized that our pricing was way too low and that the value that we were providing business customers especially was far far greater than the amount what we were charging. So we basically like doubled our pricing three times during the batch but kept on those same plans. And at first we just included a set amount of usage but then we started a credit based system but now our recent pricing model is even completely different than what we had before. >> Yeah, it’s still credits but it’s very transparent. Uh so we sell tokens at cost, we sell the compute at cost and

20:04 the APIs at cost and then we charge an orchestration fee on top of it. Very similar to open router. It just feels like if you’re going to be in this build versus buy debate to be considered a tool that is in like the build camp. Like when people say they’re going to build something internally, they’re going to use Burcell and they’re going to use Cloudflare and they’re going to use like these pieces of infrastructure that help people build. Yeah. >> But the pricing being transparent is such an important part of being in that group. Uh and we want to like most of our customers are deploying it in their own cloud. It feels like internal infrastructure and it could be part of that build camp. But if you have opaque pricing that’ll never happen like you’ll

20:34 always be the enemy. >> So we we changed our pricing recently to just be like very clear about where where GMO’s value is and what our margin is. And it’s totally agnostic then like if somebody wants to use claude or has an agreement with open AI or wants to route everything through their brain trust or their an uh Amazon bedrock proxy or something like that they can totally do that just using gum loop as well. >> What’s an investor meeting like for you today like like is it just a demo like maybe like the CDNA >> I I I learned pretty quickly that with a tool like ours showing the value with a demo is 100 times more impactful than

21:04 just like talking about what we do. Like even in this discussion now people are probably rolling their eyes when we’re like yeah we automate work and it’s so amazing. They’re like, “Yeah, everyone does.” But like if you see it working, you you start to believe it. So now I I exclusively just open the product and show people that what it can do or give them access so that they can start using it. That speaks way louder than like a pitch deck. >> You raised three rounds of funding uh well after see first round and then uh Nexus and then benchmark. You talk a little about how they all came came around and sort of like what you’ve learned about fundraising. the first

21:35 round with first round was uh the most like that’s the only time we actually fundraised. >> Yeah. >> Um we did the whole process like what YC helps you get set up to do like a week of meetings and talking to every VC. >> Extremely tiring. Uh and I think they were the best possible firm we could have went with like they’ve been amazing partners. It’s very similar to like dating I guess or I don’t know what analogy to use but building the company and gaining traction is the most important thing you can do. It’s not about like tricking an investor to think that your company’s cool or having a

22:06 better pitch deck than someone else. Yeah. Like they will always follow your company and try to invest if you’re showing that you’re changing the way people work or like having some impact in the market. So our other two rounds were preempted. >> Um we didn’t raise, we didn’t make a deck. We we just uh kept building and I think the fact that we didn’t need the money and we’re going to succeed regardless is what makes someone want to invest. >> What’s the experience of being a Canadian founder moving to San Francisco? I I remember there was like some tweet storm about about this or like advice to Canadian founders who either want to stay in Canada or people might want to move here like what’s

22:37 >> what comes to mind. >> You can absolutely stay in Canada like there’s no we have an office in Canada. We employ a ton of Canadians. I think we’re going to be going back to Canada at some point. Um so I don’t think it’s you must move to SF like a lot of people say. >> I think our approach was like we had every odd stacked against us. We still do and like we want to succeed by any means necessary. So, if there was a half a percent better chance that we might meet a customer or someone we want to hire in San Francisco because they happen to be down the street, like Instacart is was like a block away from our initial office. Yeah, >> we were going to do that. But I I think

23:08 you can 100% stay in Canada and build like a generational company there. There’s a lot of great Canadians who are comfortable and happy and they don’t want to leave the country. Um, you can hire them and you have a kind of a an advantage in that market for hiring. >> What are the best hubs or strongest like either most ambitious or best engineering talent right now? >> It’s just where more people are. Yeah. Vancouver and Toronto I think in Canada. Um >> Montreal too. >> Montreal as well. Yeah. [laughter] >> How big is the company right now? >> Between 45 and 50. Yeah. >> So like it’s interesting because there’s like the sort of people were like I’m

23:39 trying to keep the team small. You’ve been trying to do that but it’s going really well. So it seems like still you are small for the traction that you have but still now this like a big company. >> Yeah. Our initial plan was we were going to be a 10 person billion dollar company. That’s what I put out on Twitter because we raised our series A when it was Rahul and I and two interns. So, we thought like I think we’re a little cocky and thought that we can just keep it super lean. >> But then we discovered enterprise sales and [laughter] just like how much support a large company needs. Yeah. Um >> it it became obvious like the more

24:09 people we have the better. So, we’re still trying to stay super lean. And uh the the I like to refer to this multiplier where like I I’ll ask a customer who’s onboarding like how big do you think our team is? And they’re like 250 people, 300 people. And I’m like, “No, we’re only 50 or 30.” The multiplier of like how big they think we are to how big we really are is kind of what I’m optimizing for now. Um, less so like the actual objective headcount because I think we’re hamstringing ourselves a little bit by being like only 10 people. The 11th person adds a ton of value. >> What constrains success for the company right now? Is it like models are still

24:39 like not perfect the way they need to be or like like what what will be additional unlocks in the future? Do you think >> people don’t know about us? That’s uh one big thing. Sometimes like our I have calls with customers almost every single day where they’re like inbound or they heard about us through a friend and then I actually show them the tool and they’re kind of shocked at how much it can do. >> Um we’re trying to fix that so that more people know we exist and how much work we can automate. >> That’s one problem. The other one is we just don’t have enough people as ironic as that sounds based on like the small team thing. There’s like a hundred

25:10 things that our customers are asking us for that we could be building right now but we’re very thoughtful I think about who we hire and whether they’re a good fit for the company. So it just we’ve been hiring more slowly than the average company, but we we’re hiring engineers, designers, AEES, customer success. >> If I was someone who’s like never used anything like this and you’re like I’m I’m AI curious and I work for a big enterprise, what’s the vision that you would sell me on or vision that you would describe to me of like this is how you could run your company differently? Like is is there like a a way to summarize h how how a company like your

25:41 customers are compared to the ones that are not? I think the the main e like thesis is like the people who understand the tasks should be the ones automating them. Um, and the anti pattern we’ve seen over the last couple years is like you might have a a marketing ops team that has like something extremely tedious and cumbersome that’s you’re losing tons of revenue by not doing. they’re going to hire people if to solve it or they’re going to like write a spec doc of the problem they want solved with an agent or AI and they’re going to give that to an engineering team or they’re gonna give that to a consultant who’s going to charge them a crazy amount of money and they’re going to go through

26:12 this loop of like not quite right feedback meetings and you’ll get something halfbaked. >> I think like the people who really deeply understand a task should be the ones like just transferring what they understand into an agent to automate it. And the hard part is like the the biggest blocker is the a tool that they can understand. Now we’re at the point that anyone can build an agent that’s relatively solved. And then the meta problem behind it is like do they have access to the right data, the permissions, like has it let them do this sort of thing. So we’re secretly like catering to IT teams with all the features we build first is if they give the green light then anyone can automate like hundreds of hours of work and scale

26:43 their output. >> Um but I think it being on board and that comes from leadership like a a company like Shopify is so AI native that they’re going to clear the way to make sure this is possible. Whereas like a large health tech company in Europe, they’re going to have more rigidity IT team saying no to everything and you have to like work through that to get the unlock of automation. >> A lot of people that are pitching AI tools for companies are basically saying this tool will sort of like automate this work and therefore be fewer people. It sounds like you actually are describing something that’s complete opposite where everyone basically instantly have like more people and do

27:13 more. Talk about that. >> Yeah, I think part of it is like this future world in which everybody is a manager and we’re kind of all overseeing agents and like having agents do work for us. I think, you know, that seems to be what we’re trending towards. And so the way that we see it is like you don’t necessarily need less people, but the people that you have can do a lot lot more. And we see that within our engineering organization at Gum Loop as well. Like we have eight core engineers that are working on the product that includes myself and everyone’s able to do the work of like five people basically using agents. Yeah.

27:43 >> What’s the hardest challenges that you guys are facing right now? So someone who’s thinking about joining at Gum Loop um what would they work on? Yeah, I think it’s a lot of the stuff that the labs are dealing with as well and the biggest challenges the labs are facing at the application layer, which is like how do you get the most out of the agents and the models in their current form and how do you get them to do real work? How do you get them to do like long running tasks that take, you know, multiple hours? Uh how do you get them to like use computers and like browsers and efficiently and stuff like that? I think a lot of the appeal at working at a company like Gum Loop is that you’re really at the frontier and pushing the

28:15 limits of what’s possible to do with AI and the current models. Um, and I think that’s really exciting for people, especially people that are at a more kind of traditional company. >> There’s uh, every feature that people think we should build and that we start building comes with like a hundred features underneath them that are secretly the hard part. Like, yeah, browser automation is something we’re working on. Uh, but then okay, if people can automate with the browsers with agents in the cloud and on their local desktop, how does it make sure that they’re not going to certain websites or that it’s not doing anything nefarious or that they’re able to watch back uh a

28:47 session recording? And then they’ll ask like, how do I watch back 5,000 session recordings and understand what agents are doing in aggregate? How do I make sure it’s efficient from a cost perspective? Like everything that comes with those features is the hard part. Um, so we have to think of like the whole picture when we’re deploying it in a in a company, but there’s so many little edges to every feature that we need great engineers to work on. >> What’s the dynamic with the labs? Cuz you you mentioned in the past that your customers already have relationship with the labs and um like some of them have competing products. I’m assuming co-workers in this in this world as well. >> Totally. Yeah, there’s a ton of competition. I think we exist in like

29:18 this sweet spot in between the labs where um we’re kind of like the Switzerland between the labs and you can use your inference from wherever and that becomes quite appealing to like an IT leader or any leader at a company when you’re like I can build agents across the business without being locked into one of these vendors. When a better model comes out next week on Twitter and everyone’s going crazy over it, I can oneclick switch to that inference provider. And then when those companies try to approach us with like an 8 figure contract and we’re all using their solution and we can’t go anywhere, we actually have the ability to walk now to a different inference provider. And the idea that like you own your agents like

29:49 the agentic sovereignty by you deploy it in your cloud, you own the database where the traces are stored, you own the full integration stack like you’re owning your agentic future becomes quite enticing especially as yeah I think like the the model layer will get commoditized and the labs will need to generate revenue in other places and they’re going to go after their customers businesses. Uh we’ve already seen that with a few of our customers who have like the labs came up with direct competitors against them and they’re they’re kind of like burned by that experience. Most companies will probably face that in the future. So owning your agents is quite important. >> Do your customers need frontier models? >> Yeah, I think so.

30:20 >> For many of their tasks, >> there’s a they need the choice for sure because some things are so complex like you’re informing the CEO about a growth function that is like going to be in the board deck next week. you it would be better to use a the most qualified model you possibly can, but then you’re running a support use case 80,000 times an hour like using the cheapest model in the market will actually make a difference in your >> your revenue. Yeah, and I think like that’s a key thing like you kind of do want the choice because not all use cases need frontier intelligence uh even of our customers and so giving people the option to switch down and switch up

30:51 and I think having that control at the organizational level. So like admins and people who are in the know can like force people to use certain models or force certain teams and certain use cases to use certain models because if given the choice probably every individual employee is just going to use like Frontier Intelligence cuz like there’s no limit and like who cares, right? But like if you give the people that actually know what they’re doing the ability to kind of enforce different things um that we find that works out better for everyone like everyone has a better experience. It’s it’s cheaper and faster for the end user and then you know for for the company it’s also more

31:23 efficient. >> Um typically there’s one buyer for for each product category one product but it sounds like in this case could be still a bit bottom up where someone actually discovers this like how does it bubble up to the decision maker? >> It’s like a riddle every time. Uh it’s different. We’re definitely forming more of a pattern, but the person that bubbles up to and the person that gets most excited is normally like an IT leader leader like the person responsible with the AI rollout and security of the fact that everyone’s using 80 different tools and the data is going everywhere and they need to kind of wrangle all of that. Yeah. >> Um they tend to be the person we were

31:53 most excited to meet because their eyes kind of light up when they see our feature set. Yeah. >> And then they’re the ones unblocking uh the sales team to do sales related use cases and the customer success team to do their use cases because they’re giving them access to all of like the internal data. But yeah, it tends to bubble up to them and and sometimes it starts with like an individual who found us on Twitter or saw our post on LinkedIn and then they showed it to their manager, they showed it to their manager, we gave them a pilot, they showed they generated like 3x more revenue with a small sales team, for example, in a week than they did la like the three months prior. >> Can you give a pilot to them without

32:24 getting to the decision maker? >> We have to talk to it to get the pilot working, right? >> Uh that that’s like the hard part. Yeah, it’s something we’re learning and figuring out over time still, but it’s not super helpful to have a pilot uh and then they can’t connect any of their tools. So, like we kind of made >> that mistake in the past. >> Yeah, >> the number of connections that you have access to is directly correlate to how good your experience is using product. >> How important was it for you guys to do and move to San Francisco? >> I think it was pretty critical. >> Yeah, [laughter] we were uh it gave us like instant validation I think to to >> internal validation or external

32:54 validation. maybe a little bit of internal validation, but mainly external validation that like people would take our call. You know, before that we were just two guys with laptops being like, “Yeah, we’re automating work and no one cared.” Um, but it gave us like a jump start to just start building momentum and then a lot of guidance for how to like harness that momentum into something bigger. >> Uh, so it definitely was like a before and after moment for us. >> You’ve always seemed to me very self-motivated. >> Uh, tell us more about that. Tell me more about that. >> I think it’s like a fear of mediocrity. I’m just like terrified that I’ll look back in 30 years and realize like I

33:25 wasted an opportunity. Um, which >> I don’t know where that comes from, but what motivates you? >> Yeah, I think like when I was working at Amazon, uh, you know, out of university, my like goal was to get a job at a big tech company, and I was like, “Oh man, if I do this, like I’m going to be set for life.” And so like I went and did that. And within like a week, I was like, “I cannot do this for the rest of my life. Like there’s no way I can work here.” Because it was just like the things were just moving so slowly. And I felt like what we shipped with a team and 10 people, I could do with like my buddy on a weekend. you know what I mean? And so like I had to be in an environment that we were working at a fast pace and just like shipping things,

33:56 you know, quickly. Um and uh that’s that’s what kind of motivated me. Yeah. >> Awesome. Thank you so much for coming and speak to us. This is a really good conversation. >> Thank you. Appreciate it. Thanks. [music] >> [music]