In this conversation, Nataraj sits down with Jacob Bank, the founder and CEO of Relay, a startup building AI agents to revolutionize how we work. With a rich background in AI and productivity from his time at Google and as the founder of Timeful (acquired by Google), Jacob offers a unique perspective on the intersection of automation and artificial intelligence. He shares the winding journey of Relay, from its initial concept as a cross-product collaboration tool to its current form as a powerful AI agent platform. Jacob dives deep into the challenges of building in a rapidly evolving market, the importance of robust integrations, and the product-led growth strategies—particularly his success on LinkedIn—that have propelled Relay to early product-market fit. This discussion is a masterclass in modern company-building, navigating the AI landscape, and understanding the future of automated workflows.
→ Enjoy this conversation with Jacob Bank, on Spotify or Apple.
→ Subscribe to our newsletter and never miss an update.
Nataraj: What is Relay and how did it get started?
Jacob Bank: Relay.app is a platform to build AI agents. The journey has been a bit winding to get here. I started the company in 2021. My background is in academic research in AI and then building productivity tools. When I started the company in 2021, we had two founding premises. The first was that we all use a lot of tools to get our work done. The builders of those individual tools spend a lot of time figuring out how to make experiences within those tools better, but probably not enough time figuring out how those tools interact with the others in your ecosystem. For example, I used to be the product lead for Gmail and we would agonize about every single pixel when you were archiving, replying to, or starting an email. But if you said, ‘Hey, I really need to get data out of Gmail into Notion and then also Salesforce and then send a Slack message,’ we’d say, ‘Sorry, good luck. Use the API.’ When we looked at what knowledge workers are actually doing, a lot of what we do is take stuff from one tool, do some stuff with it, and then stick it in another tool. We thought there was an opportunity where people were underestimating the importance of cross-tool coordination. And second, it sounds so silly to say this now, but it was not obvious in the summer of 2021 that AI was going to be important in doing this. The original name of the company was Collab AI, and we didn’t know exactly what product we wanted to build, just that it was going to help with cross-product workflows and that it was going to use AI somehow to do that.
For the first year and a half of the company, we wandered in the desert, as they say. We built eight or nine different product prototypes that all fit that theme, but none were quite right. We built an automated to-do list, a contextual knowledge base, a stand-up tool, and an employee onboarding tool. Eventually, we landed on a workflow tool. A workflow tool that captured repeated tasks that had an element that can be automated and an element that required human judgment. That’s why we named the company Relay, because we were thinking so many things we do should be a relay race where the computer does some stuff and the user does some stuff. We announced our beta at the end of 2022 and ran it in 2023. By the end of the beta, we realized that maybe there was a category to be created there, but it wasn’t us who was going to create it. It just wasn’t right. When you’re an entrepreneur and you’re just muscling something through that’s fundamentally not right, there’s too much friction.
So in the summer of 2023, we decided we were going to build an automation tool. We would focus on the market that Zapier is the leader in: cross-product horizontal workflow automation for a non-technical audience. But we were going to try to build the modern version of it. What makes it modern is that it’s way easier to use for non-technical people, it has AI better integrated into the workflows, and it has human-in-the-loop capabilities so you can correct your AI when it gets stuff wrong. In 2024, we were an AI-powered automation product, positioning ourselves as the modern alternative to Zapier. We got to initial traction and then early product-market fit. But we realized that by positioning ourselves as an automation product, there were two major limitations. One, you limit the audience of people that think you’re the tool for them. We want to tackle the much bigger opportunity of helping every business get more work done with AI. And second, we didn’t want to be perceived as a duct tape product that only exists to temporarily glue two products together. We want to be a transformative tool. So the evolution we’re making now is transitioning from an AI automation platform for no-code workflow builders into an AI agent building platform for everyone.
Nataraj: When I see a product similar to yours, the problem is we are using so many tools and so much data is spread across different things. How are you prioritizing which tools to bring into the platform?
Jacob Bank: Right now we have about 120 native integrations. The way we think about it is that there are about 12 categories of tools that pretty much every business needs. Every business has an email client, a calendar client, a messaging tool, a CRM, an email marketing tool, an e-signature tool. There are these 12 to 15 categories that are quite ubiquitous. In each category, there are three to 20 players that have material market share. We’ve basically just tried to work our way down that list for our target audience, which skews towards small and medium-sized businesses. If you’re a regular SMB using modern tools, you’ll probably find everything you need with Relay. Zapier has 7,000 integrations; I don’t think you need 7,000. I think that’s a vanity metric. The number we need to get to is probably somewhere between 300 and 500 for the product to really feel complete. I believe integrations are skilled labor. I don’t think this is something you can just outsource. Building a really good Salesforce integration is really hard software engineering. Second, I believe agents will only be as useful as the robustness of their ability to interact with the tools you use. There are two big schools of thought: do everything in the browser or build on top of APIs. I think every serious player will need to do both, but if an API is available, that’s almost certainly a more efficient and robust way for the AI to interact with the product. We have focused entirely on robust API-based integrations.
Nataraj: What has the traction been like? Give us a little bit of insight in terms of the scale of the company right now.
Jacob Bank: We’re now at 440 paying customers. We have about 1,200 weekly active teams. That’s up from essentially zero when we launched at the very end of 2023. I’d call it early product-market fit. My personal definition of product-market fit for a product-led self-serve business is: could I go on vacation for a week and come back with more users, more customers, and more revenue? That is now true of our business. With 440 paying customers, there’s enough there to say that you’re not just a bespoke consulting shop for one or two companies.
Nataraj: What are you doing to drive this adoption further? I think you’ve done very well in terms of product-led content growth, especially on LinkedIn.
Jacob Bank: That’s super recent; we only figured that out in the past month. We’ve been building from the back of the funnel to the front of the funnel. Meaning we started with retention and depth of engagement, then moved to activation, and now have moved to working on top of funnel. In my first company, Timeful, we got 300,000 downloads the first weekend and retained none of them. I was scarred by that experience. So for this company, I would rather have 10 rock-solid, retained customers and then figure out how to bring more in. The strategy that makes sense for our kind of product has to be facilitated word of mouth, facilitated by content, community, and partnerships. I spend a lot of my time on content creation. One type is LinkedIn posts, which are teasers that illuminate a use case for an AI agent. I’ll post something like, ‘I just built a cool AI agent to synthesize insights from my customer calls.’ I’ll make a six-second GIF about it and pair that with long-form YouTube tutorials that show you how to actually build it. On LinkedIn, I now have 15,000 followers, and I’m getting about 150,000 impressions a month. Our YouTube channel will cross 10,000 views for the first time this month. It’s an order of magnitude fewer views, but super high intent. About 25% of our paying customers come from YouTube. Now that we have a robust community, many of them are writing LinkedIn posts or building templates, which helps solve the problem of people figuring out what to use a horizontal product for.
Nataraj: How come there’s no LinkedIn integration when you post so much on LinkedIn?
Jacob Bank: We’re actually in the review process from LinkedIn right now. We’re waiting for them to flip the bit to accept us, but it’s coming very soon.
Nataraj: What integrations are coming in the next couple of months?
Jacob Bank: We have a public roadmap. LinkedIn is at the top. WhatsApp is coming—that’s a really highly requested one. Xero, the accounting provider, is another. We have a few more social media integrations to build, like deepening our YouTube integration and building an Instagram integration. We need to deepen our integrations with website builders like WordPress, Squarespace, or Wix. We just have Webflow at the moment. The drumbeat of integrations will go on forever, but we really want to cluster around the use cases where we’re seeing the most traction: content creation and marketing, research use cases in sales, and general back-office operations.
Nataraj: There’s a lot of overlap with Make.com and Zapier. Do you see them as competition, and how are you differentiating?
Jacob Bank: When people are deciding between Relay and other products, they typically consider two categories. One is the traditional automation players: Zapier, Make.com, N8n. The other is the new AI agent builders: Lindy, Gumloop, Relevance. With respect to Zapier, our main differentiations are a way easier product experience for non-technical users, AI is integrated much more natively, and we have human-in-the-loop support. For the AI agent builders, they typically have good AI primitives because they’re AI-first, but they don’t typically have the depth and robustness of integrations or perfect usability. We all have our strengths and flaws, but we’re all kind of circling around the same opportunity.
Nataraj: This is your second company, and you sold the first one to Google. What didn’t work in that company, and what are you trying to avoid doing in this company?
Jacob Bank: Two big lessons from that company. The first is that we let ourselves be a little seduced by top-of-funnel numbers when we should have focused on having 10 super high-value retained users. The second was we just didn’t understand business. It was like, oh, we’ll just build a mobile app like Instagram and eventually get bought for a billion dollars or monetize it with ads. It turned out there was this other business model of subscription B2B SaaS, which totally existed in 2014, but it wasn’t in our DNA. So for this company, I wanted to make sure it provided rock-solid value, high engagement, and high retention to a small set of customers before we expanded. And second, that from the very beginning, the business model made sense.
Nataraj: You are at the edge where you might be called automation, but if you push, you’ll be called agents. How do you think these will look in two years?
Jacob Bank: I think there are three principles that are going to be really important in agent building. One, agents will need a robust and reliable mechanism to interact with all your tools, likely through APIs in the next two years. Two, agents will need to give you some sort of intermediate representation of what they plan to do before they do it and give you the ability to give feedback. And third, you need a great human-in-the-loop mechanism to correct things and help the agent learn. I think we will primarily use natural language to instruct an agent, then iterate on its plan, and then work with it to get the final output. This pre-compiled version of the flow chart is more understandable to users and will run more reliably and faster in practice.
Nataraj: What’s next in terms of scaling? Are you planning to fundraise?
Jacob Bank: We don’t feel any need to fundraise at the moment. We have plenty of runway, and our revenue is growing quite quickly. It looks like we’d be able to make it to profitability if we had to. That said, we will likely want to raise additional capital because it’s such a big opportunity. But my philosophy on company building has changed. It used to be you raise a round and triple the company size. For the kind of business we’re building—we have nine people right now—I can see us needing 12 or 15, maybe 20, but I don’t see a near-term future where we need 200. Modern company building is going to look very different.
Nataraj: This was an amazing chat. Thanks for coming on the show and sharing all your insights.
Jacob Bank: Thanks so much, it was a blast.
This conversation with Jacob Bank highlights the incredible potential of AI agents to transform business operations. His journey with Relay provides a clear roadmap for building a modern, lean, and highly effective company in the age of AI, focusing on real customer value over vanity metrics.
→ If you enjoyed this conversation with Jacob Bank, listen to the full episode here on Spotify or Apple.
→ Subscribe to ourNewsletter and never miss an update.
