Stop re-explaining your business to the machine.

Status: an idea on the roadmap. Not committed, not scheduled, and no dates.

Every team that starts using AI seriously hits the same wall: the useful context — how we actually do this, what we decided last year and why, which rules are real and which are habit — lives in people's heads and in threads nobody can find. So it gets re-explained, in a prompt, every single time.

A Knowledge Workbench would be the place that context lives once: procedures, decisions, playbooks, the domain detail that makes an answer right for your business rather than generically correct.

Why this is a workbench and not a wiki

A wiki is written for people and goes stale because nothing depends on it. The difference here is the consumer: agents read this layer as structured context, so it is exercised continuously rather than at onboarding. Content drafting, community replies, and eventually segmentation all ground themselves in the same source, which means errors in it surface quickly instead of quietly.

It also survives turnover. When the person who knew how something worked leaves, the knowledge does not leave with them.

The honest open questions

Whether it stays living is the hard part, and we do not have a settled answer: human curation, agent-proposed updates you approve, automatic capture from day-to-day work, or some mix. Getting that wrong produces a confident, out-of-date knowledge base, which is worse than none.

We would also want any answer grounded in this layer to be able to cite which version of your knowledge produced it.

This is a roadmap idea, logged because it keeps coming up. If you are already feeling the cost of re-briefing AI on your own business, that is the conversation we want.