Manifesto

Docs rot. Shipping makes it worse.

Every software company has the same quiet secret: the help center is wrong. Not everywhere. Not dramatically. Just wrong enough that nobody fully trusts it, including the people who wrote it.

It isn’t anyone’s fault. Documentation rots by design. You ship on Tuesday; the screenshots are stale by Wednesday. A button gets renamed, a flow gets simpler, a setting moves, and forty articles silently drift away from reality. The better your team is at shipping, the faster your docs decay.

So the work of keeping them true lands on whoever cares enough: a support lead late at night, a founder the night before a launch, a writer playing detective through release notes, asking “what changed, and which of my articles does it break?” It’s nobody’s full-time job, so it’s everybody’s guilt.

And customers notice. Every stale answer costs you twice: first a support ticket, then a little trust. A customer burned by wrong docs stops reading them.

The stakes just went up

For twenty years, stale docs meant an annoyed person and one more ticket. Now your documentation has a second audience: machines. Support bots answer from it. AI assistants describe your product from it. Answer engines quote it, right or wrong, to people you’ll never see in your analytics. Stale documentation used to whisper wrong answers one customer at a time. Now it broadcasts them.

Writing was never the hard part

The industry’s answer so far has been faster writing: templates, AI drafts, one-click article generators. But generating text was never the bottleneck. Staying true is the bottleneck. A help center is a living claim about how your product works today, and a claim needs checking, not a faster first draft.

What we believe

  • Documentation should be checked against the product, not against other documents. The only trustworthy way to document software is to use it: sign in, do the task, see what really happens.
  • “Verified” should come with a receipt. When, on which build, signed in as whom, and what wasn’t tested. A green tick without context is a guess.
  • Maintenance is the product. Any tool can help you write an article. The valuable thing is the boring thing: noticing what changed and keeping every article and screenshot true, release after release.
  • People should edit, not transcribe. Your team’s judgment belongs at the top of the process, deciding what’s right, not at the bottom, retaking screenshots at midnight. And their words should never be rewritten behind their backs.
  • Never guess. An AI that invents steps is worse than no AI at all. When something can’t be verified, the honest move is to say so and ask.

What we’re building

HelpCenter.AI is an agent that learns your product by using it, checks your articles against it whenever something changes, and drafts the fixes (steps and screenshots) for your team to approve. It writes new articles too: from what it sees in your product, from your walkthrough videos, and from the questions customers couldn’t find answers to. It publishes where your customers already look, and it earns autopilot one kind of change at a time.

We’ve been in this space for years, building HelpCenter.io and HelpCenter.chat, and watching great teams lose this battle to entropy. The tools kept making it easier to publish documentation. Nothing made it stay true.

That’s the part we’re automating. Docs that rot are a choice now.

If this is your fight too.

Request an audit of your help center, or just join early access. Founding terms, and a say in what we build first.

Free during early access · No app credentials needed for the first audit