Amjad Masad raises $250M at $3B valuation as Replit agent runs autonomously for hours

Summary

Replit has raised $250 million at a $3 billion valuation, led by Prism Capital with participation from a16z, Google, Y Combinator, and existing investors. CEO Amjad Masad frames the round as validation of a thesis he's held since his teens: that software creation should be accessible to anyone, not just engineers.

Agent 3: autonomous runtime, not vibe coding

The headline product accompanying the raise is Replit Agent 3, which Masad positions as a step beyond vibe coding. The benchmark he set internally was whether the agent could run for 200 minutes without human intervention. Agent 1 ran for 2 minutes; Agent 2 extended that to 20. Agent 3 is now completing sessions users report lasting 4 hours, autonomously opening browser windows to test apps, running unit tests, and reviewing its own code. The system pits Claude, GPT-4, and Gemini against each other in an adversarial multi-agent loop — the goal being a developer proxy that works while you're away from the computer entirely.

Masad is deliberate about the distinction. Vibe coding still puts users in front of an IDE. Replit's target user doesn't want that. He cites Lloyd Frink, co-founder and executive chairman of Zillow, as a Replit champion who builds on the platform between meetings on his phone, and says he's heard secondhand that Vlad Tenev prototypes applications on Replit during meetings to show engineers what to build.

Agents building agents

Buried in the Agent 3 announcement is what Masad calls a low-key launch: the agent can now build other agents. The commercial logic is that enterprises need highly domain-specific automations — an HR onboarding agent tuned to one company's idiosyncratic process, for example — that only a single internal ops person understands well enough to specify. Replit's pitch is that non-engineers can now instantiate those automations without a software team in the loop. Masad argues this starts to displace white-collar labor at scale.

Market framing

Masad draws a sharp line between the professional developer tooling market, which he sees as zero-sum, and Replit's target. The Copilot-vs-Cursor share shift is the zero-sum fight; Replit is trying to expand who makes software altogether. He puts the addressable market at roughly the size of computers themselves — every company, from a one-person operation to a 100,000-person enterprise, is bottlenecked by software, and that bottleneck is what Replit is selling against.

On cloud hyperscaler backlogs running into the hundreds of billions, Masad is candid: there may be a short-term bubble, but the long-term value of LLMs generating software is large enough that the numbers aren't implausible. He expects the technology to displace vertical SaaS point solutions and reduce the cost of internal tooling, while freeing engineers to focus on core product rather than side-quest data pipelines.

The clearest near-term commercial example he offers: a finance team bottlenecked by data engineering can prompt Replit to connect Stripe and other sources, wait 3–4 hours, and return to a working SQL environment — no engineer required.

Should you learn to code?

Masad's answer is contextual. Aspiring founders should skip learning syntax and start building on day one — a prototype in 20 minutes, a shippable product potentially within the first day. Deep computer science credentials still matter for narrow, safety-critical domains like aerospace software. For everyone else, the skill he says matters most is resourcefulness, not any specific technical discipline. He argues conventional career advice is obsolete fast enough that young people are better positioned than their parents or teachers to figure out what to learn.