AI wiki failures, the fluff to strip and the big speed-up. We were told AI could just decipher intent from large bodies of copy, but giving AI your wiki doesn’t always give it all your business context the way you need it.
That’s because it was written for humans. Written for a machine, the same knowledge works far better.
Alexandra Buys is co-founder and COO of The Delta, a venture ecosystem that started in Cape Town and now runs an 80-person operation across a Berlin startup campus. She led the rollout of Claude across the whole business and made a deliberate choice not to just point it at the existing knowledge base.
Here’s how she created docs AI can actually use…
The move: write for the machine, not the reader
A doc written for a human optimises for readability (flow, context, story. A doc written for AI optimises for retrieval) the key facts, findable fast, with nothing in the way. They’re different jobs, and the second one is what makes your AI useful.
“These documents were written for humans — there’s a lot of fluff. It just needs the key points. Structuring things for AI is far more efficient.”
How to write docs your AI can actually use
1. Strip the fluff a human would skim
Human docs are padded with things that help a reader and hinder a machine: intros, background, caveats, and the history of why a process exists. A person skims past all of it. The AI reads every word and pays for the privilege.
Cut it. Keep the key points and nothing else. If a line isn’t a fact, the AI would need to answer a real question, its weight, and weight makes retrieval slower, costlier and less accurate.
2. Keep one topic per file
A giant doc covering ten things forces the AI to process all ten to answer about one. Break it up. One small, single-purpose file per topic: how expense claims work, how you invoice in this area, what your ICP is.
Small, focused files mean the AI can pull exactly the piece it needs without dragging everything else along. Retrieval gets sharper and cheaper the more focused each file is.
3. Write down the things that are obvious to you
The most valuable context is often the stuff nobody wrote down because everyone “just knows” it: your value proposition, the language you use internally, how you actually like things done. That tacit knowledge is exactly what makes AI output sound as if it came from your business rather than a generic one.
Capture the key points, the language, and how you want things done. It’s the difference between an AI that answers correctly and one that answers as it works for you.
4. Build a purpose-built layer instead of pointing at the wiki
The instinct is to connect AI straight to your existing knowledge base because it’s already there. Alex made the opposite call: build a separate, purpose-built context layer, and be deliberate about what goes in it.
There’s a cost: when the underlying knowledge changes, you update the layer too. But a lot of what you define: how you work, your positioning, your core processes — doesn’t change that often. A lean, purpose-built layer gives you higher-quality output and controlled token cost, and it quietly forces the internal clarity of actually deciding how you do things.
The big payoff
Docs written for the machine turn your AI from an expensive guesser into something that answers fast, cheaply, and like it actually knows your business. The same questions that used to come back vague come back right.
It takes rewriting your key docs once, tighter and more focused. It pays off on every AI answer your team pulls from then on.
You might also be interested in
Want the full playbook?
This is one piece of Building Your Business’s AI Context Layer, Alex’s full masterclass inside the Founder Collab. The full session shows you the complete context-layer architecture The Delta runs on:
The full GitHub-based context layer structure (global folder plus per-team folders) and how it stays in live sync with Claude
The interview shortcut for building your first context doc from a one-hour recorded conversation
How to maintain the context layer through Claude Code with voice notes that auto-commit and broadcast changes to the team
The escalation rules pattern that tells AI when to handle something and when to loop a human in
The skills and plugins system that lets each team own its own AI tooling without breaking the rest
You’ll also get access to 40+ other masterclasses from SA founders and operators on sales, fundraising, UX, paid media and more inside The Founder Collab.
Get more SA tech and business news, tips and business-building workflows and subscribe to The Open Letter.
