Everyone's blaming the model. The model is fine.

Over the last year I've watched the same story play out on team after team. They go all-in on AI-assisted building. The first two weeks feel like flight — velocity they've never had, a demo by Friday, the sense that they've finally cracked it. Then they hit a wall. Threads balloon. The agent forgets what it decided yesterday. Every handover leaks a little more context until the output quality falls off a cliff.

And the instinct, almost every time, is the same: swap the model. Bolt on another tool. Wait for the next release that'll surely fix it.

It won't. Because you're solving the wrong problem.

AI doesn't fail because it isn't smart enough

It fails because it can't carry context across switches — and nobody designed the context it works inside.

This is the part the "just use a better model" crowd keeps missing. The frontier models are extraordinary at reasoning inside a well-defined space. What they're bad at is holding a coherent picture of what you're building and why across a dozen threads, three tools, a week of stops and starts, and a handover to a teammate who framed the problem differently.

That's not a capability gap that a bigger model closes. A more capable agent with a vague, drifting, half-remembered context just fails faster and more confidently. The bottleneck isn't intelligence. It's the scaffolding around the intelligence — and right now, for most teams, that scaffolding is nothing at all. It's whatever accreted by accident in a chat window.

This is a UX problem wearing an engineering costume

Here's the reframe worth sitting with: the brief, the workspace, the thread handover — these are interfaces. They're the surface where a human's intent meets a machine's execution. And like any interface, they can be designed well or left to chance.

We already know how to do this. It's the oldest discipline in human-centered design: figure out what someone actually needs, strip away the noise, and hand them exactly the context required to make the next good decision. We've spent decades designing that for users clicking through screens. Almost nobody is doing it deliberately for the AI now sitting in the middle of the workflow.

So the work is real, and it's specific:

  • The brief — not a wish-list dumped into a prompt, but a designed artifact that carries intent, constraints, and the "why" in a form the model can actually hold onto.
  • The workspace — the structure of how context is organized, chunked, and surfaced so nothing critical falls through the gaps between sessions.
  • The thread handover — the deliberate design of how one context becomes another when a task passes between threads, tools, or people, without bleeding out everything that made it work.

None of that is a coding problem. It's design. Someone has to do it on purpose.

Call it a Context Architect

I've started using the term because the role deserves a name. A Context Architect designs the context AI works within — so the machine, and the humans around it, actually know what they're building and why.

It sounds new. It isn't. It's briefing — the most underrated craft in the design toolkit — pointed at a new target. The best designers, strategists, and creative directors have always known that the quality of the output is set long before the work starts, in the clarity of the ask. AI just made that truth impossible to ignore, because the machine has no politeness, no ability to quietly fill the gaps with common sense. Feed it a fuzzy brief and it hands the fuzz straight back to you, at scale.

Which leads to the observation I keep coming back to:

The teams shipping with AI aren't smarter. They brief better.

They've figured out — usually by accident, occasionally by design — that the leverage isn't in the model. It's in the context they build around it. The teams still stuck are running better models on worse briefs and wondering why the magic wore off.

If your build stalled, start here

Before you rehire, re-platform, or wait for the next model drop, ask a cheaper question first: is your context actually designed, or did it just happen?

Nine times out of ten, the wall a team is staring at isn't a capability ceiling. It's an undesigned context that finally buckled under its own weight. And that's rarely a months-long rebuild. It's usually a focused, two-week fix — the kind of sprint where you redesign the brief, restructure the workspace, and put real handover discipline in place, then watch the velocity come back.

That's the kind of work I take on at Pink Cowboy: short, sharp engagements to unblock stalled AI-assisted builds and design the context your team and your tools are missing.

If that's the wall you're looking at right now, let's talk.