Replit CEO Amjad Masad pitches the 'self-driving company' run by AI agents

Replit chief executive Amjad Masad is pushing a vision he calls the "self-driving company," a business where AI agents run day-to-day operations and employees step back to handle judgment calls and strategy. Masad laid out the idea in comments reported by TechCrunch, framing it as the next step for a company that built its business on letting people write and run software from a browser. The pitch lands as coding assistants and autonomous agents move from demos into production workflows at software firms.

The concept borrows its name from autonomous vehicles, and the analogy is deliberate. Just as a self-driving car handles steering, braking and lane changes while a passenger sets the destination, Masad's version has agents executing routine work while people decide what the company should do. Replit has spent the past two years adding AI features to its coding platform, including an agent that can build and deploy apps from a prompt, which gives the company a stake in whether that handoff works in practice.

Masad is not alone in betting on agents. Executives across the industry have spent the past year promising software that can complete multi-step tasks with limited supervision, from writing code to filing support tickets. The gap between the pitch and the product remains the central question. Agents still struggle with long tasks, break when tools change, and need human checks on anything that touches money or customers, which is why most deployments today keep a person in the loop.

For Replit, the argument doubles as a product thesis. If companies are going to hand operations to agents, they need somewhere to build, test and run those agents, and Replit wants to be that environment. The company has positioned itself against larger rivals by targeting non-engineers and small teams who want to ship software without managing infrastructure. A self-driving company, in that framing, is simply the logical extension of letting more people build their own tools.

The idea also raises questions the industry has not settled. Who is accountable when an agent makes a costly mistake, and how do firms audit decisions made by systems that act on their own? Masad's answer, broadly, is that humans stay in charge of outcomes while machines handle execution, but that division of labor is easier to describe than to enforce. Regulators in the US and Europe are still working out how liability applies to autonomous software, and no clear standard exists yet.

What to watch is whether Replit ships products that match the rhetoric. The company's agent tooling is the clearest test of whether routine work can be delegated safely at small scale before anyone trusts it with a whole company's operations. If early customers report reliable results, the self-driving company moves from conference talk to a sales pitch competitors have to answer. If not, the phrase joins a long list of autonomy metaphors that outran the software.