Notes from documentation engineering work. Design decisions, tools I studied, challenges, and observations from building documentation systems in FinTech, compliance, and enterprise platforms.
An agent answers a question. Every source it drew from is correct. The answer is still wrong. Federation solved discovery. It did not solve agreement, and that is the half nobody budgets for.
Read this postSingle-sourcing solved coordination between human authors. It does nothing once a language model regenerates your source on every query. Part 3 of the Docs as Interface series: the three things you build into the source so the part that matters survives a reader you cannot see.
Part 2 of Docs as Interface. Most teams read progressive disclosure as hide complexity behind a click. That is not what it is. Here is how to apply it across …
Templates without guardrails just move the inconsistency problem one level upstream. RepoDocs AI adds structure, AI-assisted drafting workflows, and validation in one installable system for SaaS API teams.
Not a prompt wrapper. Not a fancier autocomplete. A tool that thinks the way technical writers think, before it writes a single word.
How documentation becomes an interface when AI pulls it into the product, what technical writers and UX designers are each missing, and what a converged model looks like.
I spent twenty years assembling documentation pipelines from a patchwork of tools. DocForge is an interactive prototype of what an integrated Docs-as-Code platform could look like: six modules covering the full pipeline from Markdown source to multi-format output, with DITA topic typing, Vale style linting, and CI/CD monitoring under one roof.
Why this blog exists, what I am building, and what kind of documentation work this space will cover.