Signals Inbox·August 22, 2026·AI Workflow Automation

Why did Relay shut down and join Google?

Relay did not shut down because nobody wanted it. It had real traction, but not a strong enough independent position to justify the fight ahead, just as Chrome became a much more powerful place to pursue the same AI automation idea at Google scale.

Q1What actually happened to Relay?

Relay is shutting down its AI workflow automation service, while founder Jacob Bank and several Relay colleagues are moving to Google's Chrome team.

Relay announced the closure on July 16. New signups and paid upgrades stopped immediately. Free accounts remained accessible until August 15, while paying customers were given free access until September 14. Annual subscribers are receiving prorated refunds, and Relay built export tools so customers can recover workflows, run histories, tables, sequences and MCP servers before their accounts are deleted.

A month later, Bank announced his next move: he is returning to Google as Vice President of Product for Chrome, leading product and developer relations. Several Relay employees are going with him.

The order is useful. Relay announced a shutdown rather than a migration into Google. Customers are being told to export their work and move elsewhere, and their remaining data will eventually be deleted. Google has not announced that Relay will become a Google product.

The timing is striking because Relay was still actively competing earlier this year. In April, its own comparison pages advertised more than 200 integrations and positioned Relay as an AI-native alternative to Zapier. The company had also continued adding models, MCP support, scraping tools and new agent capabilities.

So we are looking at a startup that stopped fairly abruptly while its founder moved straight into one of Google's most important AI product surfaces.

Relay's path from launch to shutdown

Q2Did Google acquire Relay?

Google has not announced an acquisition of Relay, so the evidence currently points to a team move rather than Google buying Relay as an operating company.

That distinction is easy to lose because the founder and several employees are going to the same place at the same time.

When Google bought Bank's previous startup Timeful in 2015, the transaction was publicly described as an acquisition. Timeful stopped accepting new downloads and its technology was folded into Google products. Nothing comparable has been announced for Relay.

Relay's shutdown page instead tells customers to take their data elsewhere. Subscriptions are cancelled, refunds are issued, credentials are deleted and customer content disappears after the wind-down period. There is no announcement saying Relay accounts, workflows or contracts are moving to Google.

There could still be private arrangements around intellectual property, employee packages or other assets. Neither company has disclosed enough for us to rule that out.

But saying "Google acquired Relay" currently goes beyond what we know. Relay is closing, and Google has hired Bank plus several members of the Relay team.

Functionally, it looks much closer to an acqui-hire-style outcome, even though no formal acqui-hire has been announced.

Q3Was Relay actually failing?

Relay had real users and real growth, but the company had not reached a scale that made its future as an independent horizontal platform obvious.

The clearest numbers come from Jacob Bank himself. In a February 2025 interview with Startup Project, he said Relay had 440 paying customers and about 1,200 weekly active teams, up from essentially zero when the automation product launched at the end of 2023.

Bank called that "early product-market fit."

Later evidence shows the business kept growing. An August 2025 interview-based profile of Relay described the company as having reached 1,000 paying customers. A January 2026 analysis by Flywheel, produced after speaking with Bank, also reported more than 1,000 paying customers, alongside roughly 11 million content impressions and no paid advertising.

Those figures make the "nobody wanted Relay" explanation hard to defend.

There was also evidence of genuine retention. Bank explained in the Startup Project interview that his team had deliberately worked backward through the funnel: first retention, then activation, then acquisition. He said he had reached the point where he could leave for a week and come back to more users, customers and revenue.

Still, 1,000 paying customers can mean two very different things. For a narrow, expensive enterprise application, it can support a large business. For a relatively low-priced horizontal automation platform that has to maintain hundreds of external integrations, it may still leave a huge distance to travel.

Relay had proved demand. It had not publicly proved that the economics of an independent Relay were becoming irresistible.

Q4Did Relay really have product-market fit?

Relay had early product-market fit, although Bank's own comments show that the company was still working out how broad and durable that fit could become.

We should take his wording seriously because he was unusually precise about it. In early 2025, Bank said Relay had moved from initial traction into "early product-market fit." Once users became activated, he felt good about retention and expanding their use cases. Activation itself still needed work, and the company had only recently begun attacking top-of-funnel growth systematically.

That sounds like a promising startup finding its market rather than a company that had finished the job.

Relay's positioning also kept evolving as the AI market moved. The original company experimented with several productivity concepts. Relay then became a collaborative workflow tool, moved closer to Zapier-style automation, and later reframed itself around AI agents.

Bank explained the shift openly. He felt that calling Relay a "no-code workflow automation" product made the opportunity look too narrow and risked reducing Relay to software that glued other applications together. The company wanted people to see it as a platform for building AI agents.

That repositioning helped Relay tap into a larger market, but it says something too. The company was still figuring out the strongest way to explain what it had become.

Relay had enough product-market fit to prove the product was useful. Whether it could actually own the category was much less settled.

Q5Was Relay big enough to support its business model?

Relay's roughly 1,000 paying customers were meaningful validation, but probably still small for the kind of horizontal infrastructure the company was trying to build.

Relay's published pricing shortly before the shutdown makes the scale question easier to see. The Professional plan started at $19 per month when billed annually. Team started at $59 per month for up to ten users. Larger customers could buy enterprise plans, but Relay never disclosed how much revenue came from them.

Even without making a revenue estimate, the tension is pretty obvious.

Relay wanted to work across email, calendars, CRMs, spreadsheets, project-management software, content tools, databases, social platforms and many other categories. Every new category meant more integrations, more authentication flows, more APIs to maintain and more possible failure cases.

At the same time, Relay's own documentation said most customers worked at companies with between one and 500 employees. It explicitly warned that companies with complex enterprise procurement and strict regulatory requirements might find Relay less suitable.

That put Relay in a demanding part of the software market: broad product scope, relatively low starting prices and lots of small or midsize customers.

More than 1,000 paying customers sounds impressive for a young startup. Against the cost of becoming a permanent automation layer across thousands of business applications, it starts to look much earlier.

A startup can be growing quickly and still decide that the destination is simply too far away.

Q6Was Relay's battle with Zapier getting harder?

Relay was still closing part of the gap with Zapier, but the gap that mattered most remained enormous.

Bank said in early 2025 that Relay had about 120 native integrations. He believed the company needed roughly 300 to 500 for the product to feel complete for its target customers.

By April 2026, Relay advertised more than 200 integrations. That is substantial progress in roughly a year.

Zapier kept moving too. Its current website advertises more than 9,000 app integrations, 66,000-plus triggers and actions, and usage across more than 3.4 million businesses. Zapier has also expanded aggressively into AI agents, MCP, SDK access and governance.

Bank had a fair argument that no customer needs 9,000 integrations. He called Zapier's raw integration count partly a vanity metric. But Relay still had to cover enough of the long tail that customers rarely hit a missing connector.

The numbers make the workload clear. Relay had moved from around 120 integrations to more than 200, yet even Bank's own lower bound for "complete" implied another roughly 100 integrations. The upper bound implied another 300.

And the work never really finishes. APIs change, authentication changes, permissions change and customers ask for more actions inside existing integrations.

Relay's integration gap with Zapier

Q7Did Relay's integration strategy become too expensive to scale?

Relay's obsession with high-quality native integrations produced a better product experience, but it also created a permanent engineering burden.

Bank described integrations as "skilled labor" in his 2025 Startup Project interview. His point was practical: a useful Salesforce integration needs much more than a working API call. It has to handle authentication, custom fields, permissions, objects, errors, pagination, rate limits and all the strange configurations customers have built over the years.

Relay wanted to control that experience rather than outsource it.

That makes sense when integration quality is a major product advantage. It also means growth creates more maintenance work.

Bank expected Relay to need somewhere between 300 and 500 native integrations for its target market. At the time, Relay had around 120. Even after crossing 200 integrations in 2026, the company was still building toward that threshold.

Meanwhile, every existing connector could break when another company changed an API or authentication policy. Relay's shutdown status page was still tracking incidents caused by upstream providers this year, a mundane but useful reminder of how dependent an automation platform is on software it does not control.

Bank once described the integration work as a "long slog." That assessment aged well.

Then AI opened another route. An agent that can reliably operate a website may sometimes avoid the need for a bespoke API integration altogether.

That possibility became much more credible while Relay was still investing heavily in the API-first route.

Q8Did AI agents make Relay easier to replace?

AI expanded Relay's market, but it also pushed many competitors toward the same product Relay had been building.

Relay was early to a useful combination: deterministic workflows where reliability matters, AI steps where judgment helps, and human approvals when the software should stop and ask.

That architecture remains sensible today. Fully autonomous agents still make mistakes, especially when the task involves money, permissions or ambiguous instructions.

The competitive advantage became less unique, though.

Zapier now sells AI workflows, agents, MCP access and a developer SDK on top of its existing integration network. n8n combines workflow automation with AI agents and an enormous developer community. Gumloop built an AI-native automation platform from the beginning. Lindy, Relevance AI and other companies attacked overlapping agent use cases.

Model providers also made tool use much easier. MCP spread. Computer-use models improved. Agents became better at deciding which action to take without the user drawing every step in advance.

Relay adapted to this. Its 2026 product leaned much more heavily into the idea of building an "AI team," while the assistant could generate workflows conversationally.

But the market was converging around the same insight.

Relay's early advantage came partly from understanding how to mix AI with reliable workflows. These days, that has become close to a baseline requirement for serious automation products.

AI validated Relay's thesis while making the field around it much more crowded.

Q9Did browser agents change the game for Relay?

Browser agents changed the strategic equation for Relay because they can reach software through the interface users already have, reducing the need to build a dedicated API integration for every application.

This is where Bank's comments before and after Relay become especially revealing.

In early 2025, he described two main ways agents could interact with software: operate the browser or connect through APIs. Bank believed serious agent platforms would eventually need both, but Relay had deliberately focused on robust API integrations because APIs were more efficient and reliable for the applications its customers used.

That was a reasonable decision at the time.

Google then pushed browser automation much further. Chrome introduced Gemini-powered auto browse, allowing the browser to execute multi-step tasks across websites. Google subsequently expanded Gemini in Chrome across more countries and devices.

More recently, Google integrated Gemini Spark directly with Chrome. With permission, Spark can use logged-in accounts and saved passwords to handle web errands, while returning control to the user for sensitive actions such as payments.

Compare that with Relay's old problem. Relay had to build a connector for Salesforce, another for HubSpot, another for Notion, another for Airtable, and so on. Chrome can already see and interact with the websites users have open.

APIs remain better for many high-volume, deterministic business processes, so browser control will not erase workflow platforms. But the browser suddenly covers a huge class of tasks that once required dedicated integrations.

Bank is now joining the company that owns that browser.

That's a lot more consequential than a simple change of employer.

Q10Why is Chrome a better home for Relay's idea?

Chrome gives Bank access to the distribution, context and execution layer that Relay spent years trying to assemble from scratch.

Statcounter's latest full-month data puts Chrome at about 68% of worldwide browser usage and roughly 71% on desktop. That is an extraordinary starting position for any AI agent product.

Relay had a much harder onboarding path. A user had to hear about Relay, create an account, decide what to automate, connect applications, authorize credentials and build a workflow.

Chrome is already sitting in front of the applications.

Google is also connecting the browser to its own ecosystem. Gemini in Chrome can work with Gmail, Calendar, Maps and YouTube. Auto browse can perform multi-step actions on websites. Personal Intelligence can bring more context from a user's Google activity.

The direction has accelerated lately. Google has expanded Gemini in Chrome internationally, brought auto browse to Android and connected Gemini Spark with Chrome so that longer-running agents can use logged-in browser sessions.

Bank's new role lands directly inside that buildout.

He said Chrome was "a perfect place to collaborate with agents" and described Google as an opportunity to bring what he had learned to "many, many more people."

That is probably the most literal explanation of the move.

Relay had to manufacture distribution around its automation product. Chrome already owns the place where a large share of the world does the work Relay wanted to automate.

Q11What does Google actually get from Relay's team?

Google gets a group that has spent years studying how people delegate real work to AI, led by someone who already knows Google's productivity products extremely well.

Bank originally joined Google when it acquired Timeful in 2015. He went on to work across Gmail, Google Calendar and Google Chat, and helped build AI-assisted productivity features before leaving in 2021.

Relay then gave Bank another kind of experience.

Instead of shipping AI to Google-scale audiences, the Relay team sat close to customers trying to automate messy business processes. They saw where agents failed, which integrations customers kept requesting, when users wanted human approval, how people described tasks in plain English and which processes were predictable enough to automate safely.

That knowledge maps almost perfectly onto Chrome's current problem.

Google already knew that agentic browsing mattered before hiring Bank. Chrome announced agentic capabilities in 2025 and launched auto browse before Relay shut down. Google clearly did not need Relay to discover the idea.

What Bank and his colleagues bring is product judgment built from running thousands of real workflows for customers.

The fit also goes beyond the founder. Relay was created by a team with unusually deep Google productivity experience, including former Gmail and Calendar people.

For Chrome, that background is highly relevant now that the browser is moving from displaying software toward acting inside software.

Q12Did Relay run out of money?

We have no evidence that Relay literally ran out of money, although its funding history makes financial pressure a reasonable part of the explanation.

Relay disclosed two funding rounds. Khosla Ventures led a $5 million seed round announced in 2022. Andreessen Horowitz then led another $3.1 million investment around Relay's public launch in 2023.

That gives us $8.1 million in publicly disclosed capital.

No later round was publicly announced before the shutdown.

That absence becomes more interesting when we compare Relay with competitors. AI automation attracted much more capital while Relay remained on its original funding base. Gumloop, for example, raised a $50 million Series B this year. n8n's latest strategic investment from SAP values the company at $5.2 billion, more than double its previous valuation.

None of this tells us Relay's bank balance. The company may have been capital efficient. Revenue may have covered a meaningful part of its burn. Google's offer may simply have been more attractive than another fundraising cycle.

So the narrow conclusion is the useful one: Relay was competing in an increasingly capital-intensive market without publicly raising again after 2023.

Financing pressure is credible. Cash exhaustion is not a confirmed fact.

Relay's funding position versus larger automation competitors

Q13Was Relay being squeezed by stronger competitors?

Relay was increasingly surrounded by competitors that each had a distribution advantage it lacked.

Zapier owns breadth. Today it connects more than 9,000 applications and says more than 3.4 million businesses use its products.

n8n owns a different kind of scale. When SAP invested this year, n8n reported 1.7 million monthly active builders and more than 1,400 enterprise customers. SAP is also embedding n8n directly into Joule Studio, giving the company another route into large enterprises.

Gumloop represents the AI-native challenge. Its recent $50 million Series B gives it far more capital to build, sell and support essentially the same broad promise: let companies automate work using AI.

Relay occupied an uncomfortable position between these groups.

It did not have Zapier's integration network. It did not have n8n's open developer ecosystem and enterprise footprint. It did not have Gumloop's recent funding firepower. And it obviously did not have Google's browser distribution.

Relay did have a highly regarded product, unusually good human-in-the-loop controls and a reputation for ease of use. Its reviews were strong, and Bank had built an effective founder-led growth engine.

The problem was whether those advantages could compound faster than the structural advantages around it.

By the time Relay closed, that bet looked much harder than it had two years earlier.

Q14Does Relay's shutdown mean AI workflow automation is dying?

Relay's shutdown says very little about demand for AI automation because the rest of the market is currently expanding, raising capital and attracting large strategic partners.

Zapier has built hundreds of thousands of AI agents on top of its platform. n8n says it has 1.7 million monthly active builders, more than 1,400 enterprise customers and a $5.2 billion valuation. Gumloop just raised another $50 million.

Google itself is putting agents directly inside Chrome.

The interesting change is where the value may settle.

Some automation will remain in dedicated workflow platforms because APIs offer reliability, speed and auditability. Some will move into developer-oriented orchestration systems. Some will happen directly inside AI assistants through MCP and other tools. And a growing share can happen in browsers that already have access to the websites, credentials and context involved.

That makes life harder for a standalone horizontal product without a major distribution advantage.

Relay's closure tells us more about market structure than market demand. AI automation is becoming important enough that larger platforms want to own the layer themselves.

For small independent companies, being useful is no longer enough. They increasingly need control over a distribution channel, a developer ecosystem, proprietary data, enterprise relationships or some other advantage that gets stronger as the market grows.

Relay had a strong product. Its structural advantage was much less clear.

Q15Was joining Google a good outcome for Jacob Bank?

For Bank and the Relay team, Google gives them far more leverage to pursue the product idea they were already chasing, even though we cannot tell whether the outcome was financially good for Relay's investors.

Those are two different questions.

From a venture perspective, shutting a company with more than 1,000 paying customers and moving the team to Google is obviously less impressive than building the next Zapier. No acquisition price has been announced, so we cannot judge shareholder returns.

From a product perspective, the move is much easier to understand.

Bank has spent most of his career trying to build software that proactively helps people get work done. Timeful applied that idea to scheduling. His first Google stint applied it to Gmail and Calendar. Relay expanded the idea across business software. Chrome now gives him a surface that reaches most of the web.

There is also a curious repetition in his career. Google acquired Timeful, and Bank spent years bringing its ideas into Google products. Relay has now ended without a disclosed acquisition, yet Bank and several colleagues are again moving into Google to work on a product closely related to the problem they were solving as a startup.

As of now, we cannot tell how much Relay technology will influence Chrome.

We can be much more confident about the fit between Relay's accumulated expertise and what Chrome is trying to become.

Q16So why did Relay shut down and join Google?

Relay shut down because it had found real demand but had not built a strong enough independent position to justify the increasingly difficult fight ahead, while Google offered the Relay team a much more powerful place to pursue the same basic idea.

Relay never published one definitive cause. There is no statement saying the company ran out of money, failed to raise, lost its customers or received a formal acquisition offer from Google.

But once we put the evidence together, the shape of the decision is fairly clear.

Relay had early product-market fit and more than 1,000 paying customers. At the same time, it was building a horizontal platform that required hundreds of labor-intensive integrations. Bank himself believed 300 to 500 would eventually be needed for the product to feel complete, while Relay had only recently crossed 200.

Competition was also getting heavier. Zapier now brings more than 9,000 integrations and millions of business users. n8n has 1.7 million monthly active builders and a new strategic relationship with SAP. Gumloop has raised another $50 million. Relay had disclosed only $8.1 million of funding and no new round since 2023.

Then the technical landscape moved.

Bank had originally chosen APIs because they offered agents more reliable access to software. Browser agents have improved fast enough that Chrome can now handle multi-step web tasks, use logged-in accounts, connect to Google services and hand sensitive actions back to the user. Google has continued expanding those capabilities, including the recent integration of Gemini Spark with Chrome.

That shift attacks one of Relay's hardest problems: reaching every application.

Relay spent years trying to build a universal AI layer across business software. Chrome already sits across much of the web and, according to the latest Statcounter data, handles roughly two-thirds of global browsing.

Once Google offered Bank the chance to lead Chrome product, the comparison became difficult for a small startup to win.

Relay's product had traction, but Relay had not secured the control point. Increasingly, that control point sits with the browser, the integration ecosystem or another platform already positioned between AI and the applications people use.

Google owns the biggest browser.

Relay closed, and the people who spent years learning how AI should perform work across applications are now helping Google turn that browser into an agent platform.

That is the best explanation we have for why Relay disappeared so suddenly while its thesis is becoming more relevant than ever.