Writing

On designing AI features, accessibility and product design inside growing teams.

Articles

Making AI Answers Checkable

AI answers now arrive as calculators and dashboards. A polished answer isn't a checked one. Here's what I'd put next to every number users must trust.

4 min read

Design Is Illegible to the Business. Here's What I'd Do About It.

Most design work is invisible to the business: dropped concepts, reframed requirements. Peter Merholz calls it illegibility. Here's what I'd do about it.

4 min read

Can an Agent Change Your Design System? The Harder Question Is Who May.

Checks can tell you whether an agent can change your design system. They can't tell you who has the authority to decide. Why that needs a named owner.

4 min read

Cheap to Build, Hard to Change: Keeping AI-Built Systems Legible

When AI makes building cheap, nobody plans for durability. Four questions I ask to keep AI features understandable to the people who use them.

5 min read

Write Your Design System for Two Readers: People and AI Agents

Design system docs are now read by AI agents on every task. Here's what to add, what to cut, and a quick test for your own documentation.

5 min read

Agentic BI Won't Kill the Dashboard. It Will Change Who Has to Open It.

The 'dashboards are dead' pitch from BI vendors misses what breaks when you remove the underlying view: trust, not convenience, is the real problem.

6 min read

The Hidden Accessibility Problem in Every Data-Heavy Dashboard

Most accessibility audits stop at compliance checklists. The real gap on data-heavy products is cognitive load — and fixing it helps every user, not just some.

6 min read

Production Got Faster. Judgment Didn't. The Real Design Bottleneck in 2026

AI closed the production gap. It didn't touch the harder half of the job: deciding whether what got built actually holds up against real data and real users.

6 min read

The Looks-Done Trap: What AI-Generated Prototypes Skip on Data-Heavy Products

A prompt can produce a screen that looks finished in an afternoon. On data-heavy products, 'looks finished' and 'is finished' are answered by two completely different lists of questions.

7 min read

The AI Speed Trap: Why Fast-Moving Teams Slow Down Around Month Three

AI-assisted teams move fastest in month one and get expensive by month three — not from bad code, but from small decisions nobody compared. My checkpoint system, grounded in real product work.

7 min read

The EU Accessibility Deadline Already Passed. Here's the Self-Audit I'd Run This Week.

The European Accessibility Act's deadline passed in 2025. A tiered self-audit — keyboard nav, alt text, transactional flows — for teams that haven't checked since.

5 min read

Before You Ship an AI Feature That Judges People, Design What Happens When It's Wrong

Every AI classifier has a permanent error rate. The fix isn't a better model — it's designing what happens to the person it gets wrong. My 5-question audit, from real product work.

7 min read