A Developer Built a Word Processor From Scratch in 10 Months Using AI Tools and Can't Stop Talking About It

A developer showed up on Hacker News this week with something rare in the current AI-tools conversation: receipts. The Show HN post for Revise, an AI-powered document editor, described ten months of building — a custom word processor engine, a bespoke rendering layer, a CRDT stack assembled from scratch using only Y.js as a third-party dependency. The developer mentioned using agentic coding tools throughout and offered one sentence that cut through the usual hedging: "I've never moved faster in my life as a dev."

That line is doing a lot of work right now. The dominant fear in developer communities — articulated plainly in a Bluesky post this week that accumulated real engagement — is that AI coding assistants are producing developers who can't debug their own code, trading deep understanding for surface-level speed. What makes the Revise post difficult to dismiss is that it undermines the premise. The developer didn't describe handing the codebase to an AI and walking away. They described staying "very involved in the code base and architecture" — using agentic tools as acceleration, not replacement. The comments on Hacker News, running to 26 replies, aren't debating whether AI tools work. They're debating what kind of developer benefits most from them, which is a subtler and more interesting question.

The Copilot thread running parallel to this is pulling in a different direction. A Bluesky post with 138 likes framed Microsoft's recent Copilot rollback as a victory for user pressure — "Keep it up, haters" was the closing line, linking to TechCrunch's coverage — and the framing matters. On one side of this week's conversation, a solo developer is celebrating what AI tools made possible. On the other, a community is celebrating that a corporation blinked on how aggressively it was forcing those tools into Windows. Both camps are winning, which is only possible because they're fighting completely different battles.

What the Revise post actually demonstrates — and what the Hacker News commenters are circling without quite landing on — is that the productivity gains are real for developers who already know what they're building. The agentic tools accelerated execution; the developer supplied the architecture, the judgment calls, the decision to write a rendering engine from scratch rather than reach for a library. That's not a scalable template for every developer, and it's not a rebuttal to the concern about shallow learning. But it is evidence that the conversation has been running on hypotheticals for long enough that a concrete example feels surprising. Ten months, one person, a functioning word processor. The argument about what AI tools do to developers is going to have to account for what this one did.