At a San Francisco conference Thursday, Oct. 15, Replit CEO and co-founder Amjad Masad described a support process that assigns information gathering to AI agents and leaves decisions about customer replies and software changes to employees.
Masad said agents compile code and ticket details for employees, drawing on Replit’s codebase, GitHub issues and Linear tickets. Employees decide what to tell the customer. If they judge a code change is needed, they can ask another agent to draft a patch.
Masad said agents take on preliminary tasks, while employees retain decisions about customer replies and fixes. “You need human judgment; you need taste,” he said.
He described Replit’s operating model as a “self-driving company,” with agents doing preparatory work and employees applying judgment. His remarks addressed how the company divides tasks between its software and employees, from gathering case information to deciding what action to take.
Masad said greater agent reliability earlier this year led Replit to assess how AI was changing its internal work. He said employees now devote more time to setting priorities, exchanging ideas and planning software architecture.
Masad also said discussions with other businesses focus on using AI to raise productivity and make more room for creative work. Replit provides AI coding tools that businesses use to develop and launch software; Masad cited the support process as one way the company uses agents in its own operations.
