Replit’s self-driving company puts agents inside its workflow

Agents take the first pass

Replit began examining AI’s effect on its own operations after agents became more reliable earlier this year. Masad said the company has substantially changed how it works, with agents now handling many processes and helping improve the company’s own operations.

In support, an agent checks the relevant GitHub tasks, codebase and Linear tickets before a request reaches an employee. That gives the employee context to decide what to do. If a support specialist chooses to send a fix, one click can launch another agent to deploy it.

The agents do most of the work in this cycle, but the decision to act remains with a person. Masad expects human judgment to remain important: people still need to make decisions and exercise taste.

People choose the work

Masad argues that agents are changing employees’ mix of tasks more than eliminating jobs. At Replit, employees spend more time choosing what to work on. He described more discussion and whiteboard work, alongside greater attention to architecture and system design.

That shift is the practical meaning of a self-driving company, as Masad frames it: agents take on routine work, while people direct the effort and shape what gets built.

In conversations with other companies, Replit is increasingly discussing how AI can raise productivity and make work more enjoyable. Masad said agents can take over boring tasks, leaving more room for creative work. He also said Silicon Valley has yet to offer a positive vision of the future, and that Replit hopes to help articulate one.

I think the case is more persuasive as a description of a changed workflow than as a forecast about work. The support example shows where the boundary sits today: agents gather context and carry out a deployment, while a person decides whether a fix should go out. What I’d want to know is how often that handoff needs correction, and how much employee time the system actually saves. Without that, “self-driving” describes the ambition more clearly than the results.