Something new appeared on company balance sheets in the last two years, and almost nobody is managing it. The prompt that turns a messy brief into a clean one. The workflow that classifies invoices. The automation that assembles the weekly report. The custom instructions that make the AI sound like your company rather than generic AI slop. The licences are the visible cost; this know-how is where the value lives.
These are real operating assets. And in most businesses they live in personal chat histories, private accounts, and the notes app on somebody's personal laptop.
How AI knowledge evaporates
The pattern is always the same. Someone builds something good. Colleagues hear about it and ask for it. It gets pasted into a chat, forked, modified, and within a month there are five versions and nobody knows which one works. Then the person who built the original changes roles, and the working version goes with them.
This is the single-expert trap wearing new clothes, and AI makes it worse, because AI assets change faster than documents ever did. A prompt library that was excellent in January is stale by June. Without a home, versioning and an owner, your team is permanently rebuilding what it already built.
What a home looks like
Software teams solved this problem decades ago, and the solution transfers almost unchanged. They keep shared work in repositories: one source of truth, version history, named owners, and review before changes go live. GitHub is the best-known example, and it is no longer just for developers. AI skills, prompt libraries, workflow definitions and automation configurations are all text, and text can be versioned, reviewed and owned.
You do not need your marketing team writing code. You need four habits borrowed from repository culture. One place where the current version of every shared prompt, skill and workflow lives, instead of everyone's personal copies. A named owner for each asset, so improvements have somewhere to go. Lightweight review before a change becomes the team version, so quality ratchets up instead of drifting. And history, so when a change breaks something, you can see what changed and go back.
Whether that lives in an actual GitHub repository, your project tool or a disciplined shared drive matters less than the habits. Though if your automations run on n8n or similar, a real repository is the natural home, because the workflows themselves are exportable files.
This belongs on the leadership agenda
Capturing best practice used to mean writing the process manual. In an AI-augmented business it means governing the prompts, skills and workflows the team runs on. That is layer three of our five-layer stack, shared ways of working, and in our experience it is the layer most companies skip: individual brilliance everywhere, none of it captured, compounding for nobody.
The businesses that get this right turn one person's good week into the whole team's standard practice. The ones that do not are paying for the same discoveries again and again.
If you want to see where you stand, the free assessment scores this layer directly, or bring it to a 30-minute diagnosis call.