At Cazoo, operations, product, engineering and legal could use the word “sale” and mean different moments. We were preparing to internalise logistics, and those differences were turning into a fight about what a sale constituted and who was involved.

We used event storming to lay the process out and built a service blueprint on top of it. In the room, we focused on getting to a shared understanding of terms we had been taking for granted.

Context engineering is the product discipline of deciding what a system can know, what the organisation means by it, which rules hold and how those claims change safely.

A lot of product people come to meaning and data models by way of engineering, but they are missing an important skill: information architecture.

Product leaders already own this whenever they define a customer, an activation, a risk, a permission or success. AI makes that hidden map of meaning executable. It turns ambiguity into product failure.

Roadmaps, event schemas, KPI definitions, policies and support macros all encode categories and relationships. If they disagree, the agent does not have a context-window problem. The organisation has a meaning problem.

Dave Killeen, a product leader at Pendo, put it plainly on LinkedIn: “Most agents you offer your users don't fail on the model. They fail on highly relevant and up-to-the-second personalised context.” Relevance is the thing to bring in. Context is relevance. It’s a product skill we’ve honed but now faces the machine instead of just people.

The push towards the semantic layer is real, and my bet is that semantic tech is the foundation rather than the destination. Everyone is competing for the layer, which is right for today. See it as the base of a thin slice and something more powerful becomes possible.

Meaning is built into everything, and my own work says software is an artefact of the cultures that build it. What used to be a soft interest, the territory of architects and UX people, is moving to the centre of a machine readable web.

At Cazoo, “sale” had several live answers: an order placed, payment captured, a car dispatched, a car delivered and accepted, or the return window closed. “Delivery” had the same problem. A van leaving the hub, arriving in the slot, the customer signing the handover and the car staying past the return window all mattered to different teams.

One word, several owners, several truths. The edge case of a delivered-but-returned car changed how every downstream number was counted. When a definition changes, behaviour changes even when code does not. Meaning is a dependency.