Hey there đź‘‹
We’ve made a couple of tweaks to the newsletter going forward. You’ll get to value faster because the very best of the week is now right up top. If you want to go deeper, the extra finds are waiting for you near the end.
This week we’re looking at the interface glue that decides whether “AI” feels useful or just noisy: accessible motion, a clear owner for design system exceptions, and a simple RISK check in release notes when model behaviour can shift under your feet.
Jump in 🦄 - Adam at Unicorn Club.
Get Unicorn Club in your inbox.
This week’s email edition includes Useful Extras and AI Coding Summit promo details (missing from this public post).
Subscribe: https://unicornclub.dev/
Sponsored by 20i
Peak Performance WordPress Hosting, No Compromises
Leave single-server hosting in 2015. Choose autoscaling Managed WordPress Hosting built for traffic surges, complex sites & demanding PHP workloads - without missing a beat.
If you only steal three ideas this week, steal these:
Build: Block letter-splitting animations unless screen readers pass.
Shape: Give one person design-system veto power, with clear intake criteria.
Ship: Add a “Relevant, Inclusive, Safe, Kind” (RISK) eval line to release notes when model behaviour can shift.
Top 3 this week 👇
Build: You Know What? Just Don’t Split Words into Letters ↗︎
Stop shipping animated type that screen readers spell out.
Why: Splitting words into per-letter elements can make assistive tech read only fragments, and ARIA “fixes” fail (for example aria-label is prohibited on div with the generic role).
Adopt: Default to no per-letter DOM splitting. If you must, include screen reader results for two browser/reader pairings in the PR. No proof, no merge.
Shape: Your Design System Needs an Enforcer ↗︎
Consistency dies from a thousand “small” component exceptions.
Why: Local tweaks compound into dozens of variations (Laura Klein’s carousel example), which raises maintenance cost and turns learned behaviour into a cognitive burden.
Adopt: Put an enforcer in the design system contribution path with authority to approve exceptions. Intake question: “Would this help three or more teams?”.
Ship: The new UX Toolkit: data, context, and evals ↗︎
Non-deterministic output needs a release gate, not vibes.
Why: In rollout, large models change over time and without evals we cannot know what is failing or whether a new version is safer or worse for your product.
Adopt: Add a RISK eval line to release notes for model changes. Note whether it was human review, code-based evals, or “LLM as judge”.
Full email edition includes: AI Coding Summit promo details.
Get it in your inbox.
Subscribe: https://unicornclub.dev/
Dive into more
Build: AI’s text-trap: Moving towards a more interactive future ↗︎ — Assistants should render components, not paragraphs. Pick one workflow step and make it structured UI.
Shape: Surfaces, capabilities, and extensions ↗︎ — Separates UI surfaces from underlying capabilities and extensions. A useful lens for avoiding “blank canvas” platforms.
Ship: Refactoring internal documentation in Notion ↗︎ — Shows a practical doc cleanup loop when search and Notion AI make stale pages harmful.
Ship: TBM 405: Hope, Context, and Control ↗︎ — Legibility versus local context, and how AI can push either direction. Worth checking where your reporting strips the context people need to make trade-offs.
Shape: The circular logic of our metrics ↗︎ — Calls out “intuitive” as shorthand for “familiar” and how salience-driven metrics make products “yell”.
Full email edition includes: Useful Extras.
Get it in your inbox.
Subscribe: https://unicornclub.dev/
Support the newsletter
📢 Book a Sponsorship
Thanks for reading
Want the full email edition (including Useful Extras and AI Coding Summit promo details) and next week’s issue in your inbox?
Subscribe: https://unicornclub.dev/
Adam from Unicorn Club
Connect on LinkedIn
Top comments (0)