The Manual Writes Itself

For most of software history, documentation was a promise.

You wrote the spec before the code, or you wrote it after. Either way, it was always slightly wrong. The code moved; the document stayed. The document moved; the code didn't follow. Somewhere between the two, the truth lived — and no one could quite find it.

The solution was discipline. Write better docs. Update them religiously. Hire someone whose job was to close the gap. It never fully worked. The gap was structural, not cultural. Two representations of the same thing will always drift apart.

Something has shifted.

When an AI can read your codebase and produce documentation on demand, the maintenance problem changes shape. Not because documentation got easier to write — but because it no longer needs to exist in advance. You generate it at the moment of need, from the source of truth. The lag shrinks. The drift becomes a question of code quality, not document hygiene.

The code is the spec. The manual writes itself.

This isn't just a workflow improvement. It's a change in what documentation is. It stops being a parallel artifact that must be kept in sync, and becomes a view — a rendered output, generated fresh whenever someone needs to understand the system.

The implications run deeper than convenience. It means small teams can build complex systems without the overhead of maintaining a separate layer of explanation. It means the question "is this documented?" becomes "is this readable?" — a question about code quality, not process compliance.

The promise was never the document. It was understanding. The document was just the best tool available.

Now there's a better one.