If you had to build context for an AI system to work alongside your team, where should you start?
Four categories. Three to prepare in advance, one to build as you go.
Standard documents. The reference material the work depends on. Some of it is evergreen (brand guidelines, tone of voice, ways of working). Some is project-based (statements of work, pitch documents, product specs, launch plans). The most valuable bit, and the hardest to keep, is the captured learnings: what worked, what didn't, what you'd do differently. Most teams have most of this already. The work is making it findable, current, and reachable by the AI.
Mapped processes. How the work actually moves through the team. The stages of a project, the handoffs, the decision points, what triggers what. Nobody owns the whole picture, which is why it's worth doing. The mapping surfaces gaps and workarounds already costing the team time.
Approval patterns. The signoffs, the approvals, the criteria. Usually a mix of formal policy and informal "always check with X first." Both matter. The informal layer is where AI assistants get into trouble, because it's the bit that isn't written down anywhere.
Digitised work in progress. The work generates valuable context as it happens, most of which currently disappears. Comments from proofing tools. Meeting transcripts. Slack threads from project channels. Files organised so the AI can find them. Start small. One project, one folder, AI pointed at it. Iterate from there.
The teams getting ahead on AI aren't doing more ambitious AI projects. They're doing the unglamorous work first. The byproduct is an organisation that knows itself.
Process is one of the five dimensions the diagnostic covers, alongside Mandate, People, Infrastructure, and Governance.
The documentation work has always been worth doing. AI just gives us the reason to actually do it.