Vision

UX in the AI era

Every design leader has a slide about AI. Fewer have shipped an answer to it. This page is what I believe about where our discipline is going, and under each belief, what I've actually built or changed because of it.

One belief sits under all of these. For twenty years my work has put a person next to a machine that can do something impressive, first a robot, now an AI agent. The machine keeps changing. The person beside it, the one who has to trust it and defend the result, is still the whole job.

1. AI features earn their place through user validation, not hype

The cheapest thing in the world right now is an AI feature. The expensive thing is an AI feature users trust and adopt. The failure mode I organize against is the AI checkbox: shipping AI because the roadmap said so.

My answer is structural, not rhetorical. I repositioned the research team as the AI validation gate. Researchers went from under 1% of their time on AI to 64% in a single year, and validation workshops became the standard checkpoint before AI features advance. Nine AI products went through deep user validation; several were reshaped or stopped by what we learned. In a regulated industry that discipline is not a tax on speed. It is what makes an AI feature credible enough to ship at all.

→ Case study: The AI validation gate · the essay: The assistant is scaffolding

2. AI UX has to move beyond chat

A chat box bolted onto a product is not an AI experience. It is an escape hatch from designing one. The real work is task flows with guardrails, provenance users can inspect, confidence communicated honestly, and explainability that survives contact with regulated users, people whose jobs depend on being able to defend an answer's source.

That is why the next phase of our design system is an AI pattern library: reusable patterns for evidence, versioning, limits, and workflow aids, so every product team starts from the hard-won answers instead of rediscovering the problems.

→ Case study: The AI validation gate · the essays: Design for the user who has to defend the answer and The door where the designer gets in

3. Pave the paths, hold the taste, connect the dots

AI is redistributing the work of design, and I run my organization on a three-part reading of it. First, everyone designs the simple. Product managers, stakeholders and engineers now get to an 80% answer with AI in an afternoon, and the design org's job is to pave the road they are already driving on: design language, templates, guardrails baked into the tools. Enable, don't gatekeep. AI accelerates development, and without a shared design language it accelerates inconsistency just as fast; the paved path is what turns that speed into a cohesive product instead of fragments.

Second, the complex still comes to UX. On the bets that matter, where user insight and taste decide the outcome, everyone still turns to the design team. AI lowers the bar; the ceiling stays ours. Every hour AI shaves off mock-ups and routine validation gets reinvested in the work only taste and research can do.

Third, the net-new craft is systems thinking: connecting products, prototypes and workflows, and spotting ideas before they are legible to anyone else. I plan for 30 to 40% of a designer's week being spent pairing and connecting across teams rather than mocking and validating, and I shape roles around that.

→ Case study: Helix, the paved path already in production · the essay: AI lowers the bar; the ceiling stays ours

4. AI transforms the practice, not just the product

Most design leaders ask what AI means for their users. The equally important question is what it means for the design organization itself. Research demand in my portfolio doubled in a year, and no research team scales linearly with that. Leverage has to come before headcount.

So I build the leverage myself. Interrogable synthetic personas grounded in real participant research, with provenance back to source transcripts and an evaluation suite guarding the grounding. Automated qualitative synthesis. An intelligence studio that lets anyone interrogate voice-of-customer and win/loss data in plain language. The rule underneath all of it: AI amplifies research; it never replaces the contact with real users that grounds it.

I have been early on this before it was comfortable. Months before agent-accessible data became my company's strategic direction, I was building it in the open: 50+ MCP servers connecting AI agents to pharmaceutical and biomedical data.

→ Case studies: Synthetic personas · Prototyping the strategy · The lab · the essays: When the method becomes code · Knowledge that compounds

5. Teams will be humans and agents, and that is a design problem

The next organizational question is not whether AI agents join our teams. It is how: with what identity, what scope, and what accountability. I run this experiment on my own infrastructure, a self-hosted workspace (built on block/buzz) where human collaborators and AI agents share the same rooms. Each agent has its own keys, its own memberships, its own audit trail, and, critically, there is an independent fact-checker agent that verifies what the others claim. Scoped identity and adversarial checking, not blind trust.

The way agents show up in a team is a UX problem, and design leaders should be the ones with a working opinion on it. I have been working on this question longer than the industry has had it. My PhD put robot swarms alongside firefighters searching buildings: human and machine teammates, trust, and who is accountable when the machine speaks. Twenty years later, the teammates became software.

→ The lab · the essay: The org chart will have agents on it

6. The design org of the future is three-legged

I structure UX as one discipline with three legs: product design, UX engineering, and research, sharing one infrastructure of design tokens, patterns, evaluation methods, and data. Not because it is fashionable, but because every hard problem above lands between the legs. An AI pattern library is design plus engineering. Synthetic personas are research plus engineering. A validation gate is research plus design. An org chart that separates them turns every real problem into a coordination problem.

→ Playbook: how the org is built

These are not predictions. They are commitments, each one running in production, in my org or on my own servers.