An Engineer Built a Feature Without Opening Her Code Editor
The 200-message thread stayed public, so coworkers could watch, join in, and copy the method.

Y Combinator · The Rise Of AI CoworkersLauren Tan, an engineer at the startup Grockbot, started with a video clip. She pasted a link she had found on X, the social media site, into Slack, the workplace chat app, and asked her artificial intelligence bot what it would take to support a feature like that.
Her bot, an engineering helper called Potato, did not answer in words alone. It started a coding assistant running on its own computer in the cloud, built a prototype inside the Grockbot app, and posted screenshots and short videos back into the thread.
What followed was a whole product build inside chat. Over about 200 messages, Tan, her colleague Samantha, and the bot designed, revised, and finished the feature, without Tan opening her coding editor or reading the formal code proposal.
Cursor is the editor Tan usually uses to write and edit code, and a pull request is the packaged set of changes a teammate reviews. Tan skipped both steps for this project.
“I didn’t even look at Cursor for this project,” she said. “We did it all in the Slack thread.”
Because the thread lived in a public channel, anyone interested could join, react to the screenshots and videos, and offer feedback. Tan said that shared setup shortened the usual loop, where she might build alone and hear reactions much later.
Tan said she, Samantha, and another product manager, who helps decide what to build, were happy with the result. She said a feature built this way should ship much faster than one made through a longer, private process.
One participant in the conversation said he had seen the teaching effect time and time again in his own work. Someone shares an idea in a feedback thread, the group spins off a bot, and the next day someone who watched tries the same move, a pattern he expects to spread beyond engineering.
Tan, who once would have done this kind of coding work alone in her editor, built the feature without ever leaving the conversation.
Questions people ask
How can it be safe to ship code that no human read?
Teams rely on automated checks that run before code can ship. Those include tests that run the program, tools that scan for known mistakes, and rules that enforce the house style, plus a trial run on a separate computer.
What is the cloud computer the bot used?
It is a rented computer in a data center that the bot controls. It can install the app, run it, take screenshots, and record video, then report back in chat. That lets the human judge the working result instead of the written code.
Why does it matter that the work happened in public?
In a private chat, only one person learns how to steer the bot. In a public thread, others see the prompts, the screenshots, the fixes, and the final result. That makes it easier for the next person to repeat the workflow.
