Relay.app

» see relay.app as it lived, on the wayback machine

~ the obituary ~

Relay.app was a workflow automation tool in the shape Zapier had made familiar: a trigger in one service, an action in another, steps in between. The part that was not familiar was the pause. Relay's signature step was human approval, where an automated chain would assemble a draft reply or a proposed update and then stop, waiting for a person to look at it and press a button before anything left the building.

Jacob Bank built it. He had sold his previous company, the scheduling app Timeful, to Google in 2015, then spent over six years inside Google as product lead on Gmail and Calendar. He left in July 2021, took some former colleagues with him, and spent two years in development before Relay came out of beta in 2023 with Zapier named as the target.

It raised $8.1 million, with Khosla Ventures and Andreessen Horowitz on the cap table. For a category where the incumbent was a decade old and enormous, that is a modest amount of money and a reasonable amount of attention.

~ the case file ~

~ the thesis ~

The approval step was a genuine product idea and it was pitched at a real anxiety. In 2023, if you wired a language model into a chain that could email customers or edit records, the honest answer to "what happens when it gets one wrong" was that it got one wrong and the email went out. Relay's answer was that it did not go out until somebody said so, which made the whole apparatus safe enough to actually put in front of a business.

Three years later that anxiety had been priced differently. The industry spent 2025 and 2026 selling agents that run unattended, measured on how few interventions they need, and a human checkpoint became something to engineer away rather than to advertise. Relay's distinguishing feature turned into the thing its competitors described as friction.

Automation tooling of this sort has never been near my work. The things I build run on a device in the air with nobody available to approve anything, so an approval step is not a safety feature there, it is a design failure. In an office the pause exists so that a named person signs off on what goes out, which is a requirement I have never had to build around.

~ the team google wanted ~

In July 2026 Bank announced the shutdown, and in August he said where he was going: back to Google, as VP of Product for Chrome, leading product and developer relations, with the stated goal of helping people "work with AI in Chrome to get things done." Several staff went with him. Google hired the people and took neither the product nor the brand, which is why there was nothing to transfer and no acquirer to thank.

Free accounts lost access on 15 August 2026. Paid accounts ran until 14 September. The page at relay.app is now four sentences long and does not soften anything: "All accounts, workflows, and run history have been permanently deleted. Thank you for building with us." Underneath that it links to the Internet Archive's copy of the July 2026 homepage, which is an unusually self-aware thing for a dead company to do, pointing visitors at the archive because it knows it is the only place the product still exists.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »