The work took about a year and covers 600+ icons in two sizes and two styles, every single exploration and every final icon variant drawn by hand.

I thought AI was about bringing engineering and design closer together, but in the end it was a QA holding the keys to a self-improving tool.

These are some of the books that we frequently recommend to designers who are setting out to make a positive difference in the world.

A standard format for describing the behavior an AI agent is expected to follow across repeated interactions. Written for the people and agents who review traces, design evals, and align prompts.

Animate Lucide, Tabler, Heroicons or any stroke icon set. Optimal rotation solved in closed form, spring physics, zero dependencies.

Users don't form first impressions based on output quality alone. Their experience begins from the moment they open the agent, and it's important that we treat early interactions with just as much care as what comes after.

Agents, tokens, context windows, MCP: the words fly around every AI conversation at work, and they make the whole subject feel more technical than it is.

This is the bit I think a lot of founders underestimate. They compare their product with the incumbent and think: ours is clearly better.

A small library of extremely crafted, copy-paste components for chat agents, thinking states, human-in-the-loop approvals, and everything agents need to talk to humans beautifully.

Grade first appeared to make type look as intended with low-quality printing methods or on different paper types.

Morph merges, reshapes and dissolves touching pieces; Move trails a moving element as liquid rubber.

Software quality now depends on the constraints you set around your agents.

A study of the current state of the art in how design systems implement and distribute agentic tooling for DS contribution and consumption.

The better you are at moving fast the worse you can be at prioritization.

The bad decision was 'just style a div and call it a button'; the overcorrection was 'abstract so hard that we've rebuilt the button from scratch, in userland, forever.'

The goal is pretty simple - start with one base colour and let the rest of the colour scheme fall out from it.

Before AI, writing code was expensive, so the cost of building imposed a higher threshold on which ideas survived long enough to be built.

What I learned building an 8-agent pipeline that produces production design system components from Figma.

How our design team stopped mocking things up and started building them — inside the real repo, with a git branch and zero prior GitHub experience.

The value is hard to pin down because the measurement system was never built to see it.

This is the exact method how to get web design clients (or any dream client) without cold calls, ads, or an audience: a love letter.