Patterns
The pattern archive.
This is where our decodes compound: every pattern we name, take apart, and rebuild lives here permanently — searchable, evidence-graded, and free. Instagram is where you'll spot the work; this archive is where it stays useful.
The archive is new
First decodes land soon. We'd rather show you an honest empty shelf than fill it with placeholder posts — everything that appears here will have passed our evidence standard first.
Until then, take the tool we grade with: The AI Interface Reality Checklist (PDF) — two axes, five slop markers, six principles.
Want to know the moment the first one ships? The newsletter is the fastest route.
The formats
How to read this archive when it fills.
Everything we publish arrives in one of three formats. Each has its own laws — knowing them tells you exactly what you're getting before you read a word.
The spine
Pattern Lab
Our core program: we decode and prototype the AI interface patterns emerging beyond chat — approval gates, thinking states, agent handoffs, uncertainty displays — plus teardowns of agent UX. Each decode names the pattern, walks its state space (trigger, in-progress, confirmation, failure, recovery), and explains the design decisions underneath.
- Every pattern post ships or reuses a built artifact — one prototype, many posts, never a re-crop.
- Follow-up posts must add new design thinking: a new state, a new interaction, or a trade-off not shown before.
- This is the format behind the 30-day promise — the vocabulary and frameworks come from here.
The transformation
Slop Makeover
We de-slop an AI-generated screen — every decision explained. "Slop" is defined behaviorally, never as an aesthetic we happen to dislike. A screen qualifies only if at least two of these markers are present, and the makeover must visibly repair every marker it names:
- Generic hierarchy — the layout communicates nothing about what matters.
- Fake affordances — elements that look interactive or intelligent and aren't.
- Inadequate state design — no loading, uncertainty, failure, or recovery states where the interaction needs them.
- Inaccessible presentation — contrast, legibility, motion, or structure failures.
- Decorative AI clichés — sparkle glyphs and gradient-orb "magic" that signal AI without informing anyone.
Makeovers ship only at our top internal grade — a below-grade week is a skipped week, not a published one. We critique interfaces and their constraints, never their makers, and we never use the format to shame a small team or an individual. Makeovers can't be bought: payment can neither place a product into this format nor keep it out.
The spotlight
In the Wild
Genuinely strong, newly shipped AI UI patterns spotted in real products — always first-hand tested by us, never someone else's screenshot, and always tied back to our framework: we name the pattern and the design decision that makes it work.
- Original analysis in every piece — never bare curation, never a raw repost.
- The product (and designer, where known) is always credited.
- Never sponsored — a paid spotlight would be an endorsement for sale, so this format carries no sponsorship, ever.
- Every spotlight feeds the spine: a pattern worth spotlighting enters our decode queue.
Spotted something strong in the wild yourself? Send it through the contact form — we test everything first-hand before it ever appears here.
Whatever the format, every claim carries a Reality Ledger record — what state it's in, and how we know. The checklist below is the same tool we grade with.