Docs in the AI Era: Keep It Simple
AI makes team docs easy to write, but they still go stale. A simple doc practice: write for the right reader, keep strict rules for machines, keep docs alive.
A doc that the right person reads and uses has value. Other docs do not.
The story
Everyone agrees that docs matter. But who updated one this week?
Across the industry, many companies fail at this. At a knowledge-management study group, I heard a name for it: the document graveyard. People write docs, and nobody reads or updates them.
My team also ran an experiment. An AI bot wrote one review file for each pull request and saved it in the repo. Four months later, we deleted 80 of those files. AI made docs easy to write, but the docs still went bad.
Why docs always lose
On the Eisenhower matrix, docs sit in the “important, not urgent” box. So every deadline wins over them.
Docs do not bring in money directly, so nobody rewards them. Lazy people are not the cause. The structure is.
The good news: AI frees up time for exactly this box. I wrote about that in The Golden Age of Quadrant Two.
Who this is for
- Senior engineers who want their team docs to stop going stale
- Teams that are adding AI agents to their development workflow
- People who must clean up a team wiki and do not know where to start
Route map
1: Why it is hard
- Write Only What Your Reader Cannot Find: an AI agent is a reader too. Write only what your reader cannot find.
- How a Doc Goes Stale: 80 review files that AI wrote, all deleted four months later.
- Name the Pain Before the Tool: docs go bad for three reasons, and AI fixes only one.
2: Our own rules
- Strict Rules Only Where a Machine Checks: if a machine can check a rule, make it strict. The rest are habits.
- A Good Name Is a Doc You Never Write: names and small symbols are docs too.
- Remember Where, Not What: remember the map, not every street.
3: Keep it alive
- Keep Docs Alive: Three Reminders, One Direction: three reminders, and changes that flow in one direction.
What you will take away
The summary at the end of the series gathers it in one place:
- the golden circle behind the series: why, how, and what
- six questions to start your team’s doc discussion
- one first step to try for a month
Start with part 01, Write Only What Your Reader Cannot Find. The series ends with a summary of what to take back to your team.
References
- Eisenhower method (Wikipedia)
- Docsie: the documentation graveyard problem
- DIKW pyramid (Wikipedia)
- TencentDB Agent Memory: the four memory layers
- Diátaxis: the four doc types and their two axes
- Bullet Journal: symbols, index, and the monthly migration
- FastAPI: API docs built from code
- GitLab handbook: how GitLab uses its handbook
- Software Engineering at Google, chapter 10: Documentation (g3doc)
- Elasticsearch: fuzzy query
Related: this series continues the AI era story from Will AI Take My Job? Same Work, More Return.