Last week I linked Ravi Mehta on the three layers of context engineering for AI prototyping: functional spec, visual wireframe, structured data. Karo Zieminski, an AI PM writing Product with Attitude, makes the same case at the product scale and cites Mehta directly. Mehta wrote about prototyping one screen; Zieminski writes about designing the whole product around an agent.

Zieminski puts it in one line:

Prompt engineering is deciding what and how to ask the model. Context engineering is deciding what the model knows when it answers.

Then the asymmetry:

A well-crafted prompt in a poorly engineered context still fails. A poorly crafted prompt in a well-engineered context often succeeds.

That asymmetry is the argument for treating context as the underlying system.

If that asymmetry is real—and a year of using these tools tells me it is—then most teams are still optimizing the wrong layer. The visible artifact is the prompt. The work that actually decides the output is everything around it.

The piece I want to underline is who owns the work:

PMs define what goes in each context layer. Engineers build the infrastructure to fetch and store it… If the PM isn’t doing this, one of two things happens. Either an engineer makes the product decision by default, or nobody makes it and the agent gets every available signal dumped into the window.

Zieminski calls the alternative abdication. I think she’s right and I also think most PM job descriptions in 2026 haven’t caught up. The hiring filter still selects for ticket-shaping and roadmap maintenance, not for “decide what the model should know about the user, what should age out, what should never get re-fetched.” Those are product decisions about how memory is organized, and the people best positioned to make them—PMs who understand the product and the user—are often the ones least equipped to talk about retrieval and eviction. The gap is one of vocabulary and authority.

Both write for PMs, but the work is also design work. The context an agent sees is a designed surface: what gets included, what gets hidden, what should age out, what should persist between sessions. Mehta’s three-layer brief—spec, wireframe, JSON, twenty minutes in Figma, real data—is daily prototyping for designers working with agents now. Zieminski’s architecture is the system those prototypes live inside. If designers don’t show up here, PMs and engineers will design this surface for us.