My team, like many, is implementing AI in our product lifecycle. I’m currently researching how best to transform our design function to make the most of these tools. As I weigh the benefits of tools like Claude Code for design, I’ve recognized some patterns worth sharing.
Not every part of design is equally automatable.
The design process has sequenced phases. At one end is production work — resizing icons, applying design system components, generating variants, exporting assets, writing boilerplate code. At the other end is strategy work — deciding what problem to solve, understanding who you’re solving it for, defining what success looks like.
AI is extraordinarily good at the production end. It’s genuinely limited at the strategy end. The design leaders who thrive over the next decade will be the ones who automate the right layers and protect the ones that matter.
The reason AI struggles at the strategy end isn’t technical. It’s that strategy work depends on context AI doesn’t have: the political constraints at your company, the relationship history with a specific customer, the memory of what’s been tried and failed before, the embodied sense of what users actually need. AI is trained on general patterns. Good design decisions come from specific contexts. The gap doesn’t close by making the model bigger.
Here’s my rough map:
Automate aggressively: Production, format conversion, basic component variants, boilerplate code, documentation drafts, meeting notes. Anything that’s repetitive, pattern-based, and where quality is checkable at a glance.
Automate carefully: Fit and finish, low- to mid-fidelity prototyping (especially with tools like Claude Code that can iterate on your actual codebase, accessibility audits, content drafts, user interview synthesis). AI can accelerate these, but a human needs to review the output critically. The failure modes are subtle enough to miss.
Don’t automate: Problem framing, user research in the field, success-metric definition, prioritization against business strategy, critique of your own team’s work, honest conversations with stakeholders. These require human judgment, context, and skin in the game that AI cannot provide.
The temptation is to push AI further up the stack than it belongs. Teams get excited about using AI to generate ideas, synthesize research, or write strategy decks. These uses can look impressive. They can trick firms into believing good design is happening when it isn’t. A roadmap deck that reads well is not the same as a roadmap that reflects real understanding of users, business, and timing. AI gives you the former. It cannot give you the latter. The polish hides the failure, and the shipped work has no impact.