Replit CEO Amjad Masad thinks the next big shift in work isn’t about job losses. He thinks it’s about what people spend their days doing. Speaking at TechCrunch Disrupt in San Francisco, Masad described what he calls the “self-driving company,” according to TechCrunch AI. In his version, AI agents handle most of the grunt work and humans add the judgment and taste.
He’s not describing a far-off idea. He says it’s how Replit already runs.
🧭 What a “Self-Driving Company” Looks Like
Masad said Replit “entirely changed how we operate” once AI agents got more reliable earlier this year. His example was customer support:
- A support ticket comes in and an agent reads it first.
- The agent triages it by checking GitHub issues, the codebase and Linear tickets.
- A human gets the ticket with all that context already gathered.
- If the human decides a fix is needed, deploying another agent to write the patch takes one click.
“You have an entire loop where agents did most of the work, but crucially, there’s human judgment in the middle, and we don’t think that’s going away anytime soon,” Masad said.
The pattern is worth noticing. Agents don’t own the workflow from start to finish. They do the research and the execution, and a person makes the decision in between. That’s a far more realistic design than the fully autonomous agents many vendors were pitching a year ago.
🔄 The Work Moves Up the Stack
Masad’s bigger claim is that agents change what people work on. “We’re spending a lot more time figuring out what to work on,” he said. He described more whiteboarding, more discussion and more time spent on architecture.
He also said AI makes work “less boring and more creative,” and that his conversations with other businesses now focus on productivity and happiness at work. “There’s a positive vision of the world that Silicon Valley is not really putting out, and we’re hoping to do that,” he added.
🤔 Read This With Some Healthy Skepticism
It’s an optimistic story, and Masad has a reason to tell it. Replit sells AI coding agents, so a world where every company runs on agents is a world that buys more of his product. That doesn’t make him wrong, but his pitch is a prediction, not proof.
There are also a few gaps in the picture:
- Replit is an unusual test case. It’s a software company full of engineers who build agent tooling. A logistics firm or a hospital system will have a much harder time making the switch.
- “Changing jobs” can still mean fewer jobs. If agents do most of the triage, a support team may not need to grow as fast. The people who stay may do more interesting work, but headcount math still matters.
- Taste is hard to measure. Saying humans will be valued for judgment sounds great. Few companies know how to hire for it, train it or pay for it yet.
Masad even admitted that much of America “still don’t know that that change is coming, and it’s going to be surprising to a lot of people.” That surprise is exactly where the backlash usually starts.
🔮 Where This Goes in the Next 1-3 Years
Here’s what I expect to happen.
Human-in-the-loop becomes the default design. Agent-first pipelines with a human checkpoint will spread from engineering to support, ops, finance and sales. Fully autonomous agents will stay limited to low-risk tasks.
Job descriptions get rewritten. Roles will shift toward reviewing, deciding and setting direction. Entry-level work that used to teach those skills will shrink, which creates a real training problem for companies.
Context becomes the moat. Replit’s agent works because it can see the codebase, the issue tracker and the ticket history. Companies with clean, connected internal data will get far more out of agents than companies stuck with scattered tools.
✅ What You Should Do Now
- Map one workflow end to end. Pick something repetitive, like support triage, and figure out which steps are research, which are decisions and which are execution.
- Hand agents the research and execution first. Keep the decision with a person.
- Connect your systems. Agents are only as useful as the context they can reach.
- Invest in judgment. Start defining what good taste looks like on your team, because that’s becoming the skill people get paid for.
Masad’s self-driving company is still a vendor’s vision, but the workflow underneath it is practical and easy to copy. Companies that start redesigning their loops now will be the ones setting the pace in 2028. Full coverage of his Disrupt talk is available at TechCrunch AI.