Context is the moat, not the model
Everyone has the same models. Almost nobody has their own product's memory in a usable shape.
The frontier model your team uses is available to your competitor at the same price, on the same day. Whatever advantage it confers is shared by definition. The thing that is not shared is what your organisation knows about your product.
Why that knowledge is usually unusable
It exists. It is simply not in a shape anything can read. It lives in a call recording nobody transcribed, a Slack thread that was never pinned, the third revision of a document whose first two revisions explain the constraint, and in the heads of two people who are in a meeting right now.
- Scattered across tools that do not talk to each other
- Undated, so you cannot tell what superseded what
- Unattributed, so you cannot tell who is allowed to overrule it
- Contradictory, and nothing surfaces the contradiction
The work is the shaping
Getting an assistant to produce a document is trivial. Getting it to produce a document that agrees with what your team decided last quarter requires that the decision from last quarter be findable, attributable and current. That is not a model problem. It is an archive problem wearing a model's clothes.
Which is good news, because archives compound. Models reset with every release.
Stop writing PRDs without understanding the product.
Anchor reads your interviews, tickets and docs first — then writes, citing every line.
