The Context Ceiling

There's a version of this conversation where I tell you AI has no place in product design. That would be wrong. There's another where I tell you AI changes everything and traditional methods are dead weight. Also wrong.

The truth is less satisfying: AI workflows are genuinely useful in product design, and they hit a hard ceiling exactly where the most important decisions live.

What AI is actually good at

I use AI tools constantly. Not as a gimmick — because they save me real time on work that used to be tedious.

Synthesizing research notes into patterns. Exploring visual directions before committing. Generating copy variations. Stress-testing flows by asking “what happens when the user does X” across a dozen scenarios. Rubber-ducking architectural decisions at 11pm when the team is heads-down.

These compress hours into minutes on real parts of the design process. Any designer dismissing AI outright is leaving value on the table.

But all of these tasks are generalizable. They work on information that can be articulated and processed. They operate on the surface of design — the layer made of text, pixels, and logic. The layer underneath is where things get complicated.

The context ceiling

An existing product isn't a blank canvas. It's an archaeological site.

Every screen has history. Every flow has a reason, and that reason is almost never “someone designed it this way because it was optimal.” It's because a key client demanded a specific workflow. Because a backend constraint made the ideal solution impossible and everyone forgot it was a compromise. Because the previous designer left and the engineer who maintained the feature guessed at intent. Because the compliance team has an interpretation of a regulation nobody wants to challenge.

None of this lives in a document. It lives in people's heads, in Slack threads from 2022, in the institutional memory of a team that knows where the bodies are buried.

When you ask an AI to help redesign a feature, it has access to none of this. It will produce something plausible — maybe even something that looks better than what exists. But “better” without context is just “different with more confidence.”

The plausibility problem

This is what actually worries me: AI makes uninformed work look indistinguishable from informed work.

Before AI, when a designer didn't understand the domain, you could tell. The designs were vague. The flows had gaps. The terminology was wrong. The lack of understanding was visible, which meant it could be caught early.

Now, a designer can feed an AI enough context to produce something that uses the right vocabulary and presents a coherent-looking solution. In a review, it holds up. People nod.

Then it ships and breaks. Because the solution didn't account for the edge case only a domain expert would know about. Or it restructured a workflow designed around a regulatory requirement. Or it optimized for onboarding in a product where 95% of the value comes from power users who expect continuity.

Plausible and correct are almost indistinguishable in a review. They're completely different in production.

Domain knowledge isn't a prompt

There's a belief that you can bridge the gap by giving AI more information — better prompts, uploaded requirements, research transcripts. This helps, but its limits are fundamental, not technical.

Domain knowledge isn't a document. It's a way of thinking that develops over time. After months on a product, I don't just know how it works. I know which features are load-bearing — minor-looking but untouchable. I know which stakeholder's feedback is a proxy for a concern they won't state. I know which design directions will make engineering groan.

You can't upload intuition. You can't prompt your way into political awareness.

AI helps you work faster within a domain you understand. It cannot help you understand the domain.

Where this leaves us

AI workflows belong in product design as accelerators, not substitutes for embedded knowledge. Use them for the mechanical parts. Recognize the ceiling for everything else.

The risk isn't that AI will replace product designers. It's that it will make it easier to skip the deep, slow, context-building work — and produce designs that look right but aren't.

And if you've read the rest of this series: looking right and being right are two completely different things. AI just made the gap harder to see.