Helping Humans and Computers
Understand Each Other

Design for things that have to hold together at scale. Prototypes, design systems, content architecture, and the governance that keeps them from drifting apart once more than a few people are contributing. We came out of enterprise, so most of what we do involves a lot of pages, a lot of contributors, or both.

A prototype that would survive contact with the real world. Not more pixels. It means the thing is built with real structure underneath, so a screen reader can read it, a search engine can index it, and an AI agent can understand it, and you know that before you commit a quarter to building the production version.

It's less work than it sounds, because the parts that make a page readable are mostly things you'd have to do eventually anyway. Semantic HTML, headings in order, labels on inputs, alt text. Doing it in the prototype means you find the problems when they cost an afternoon instead of a sprint.

Because a lot of your visitors are machines now. Cloudflare's traffic data crossed over in June 2026, with roughly 57% of requests to web pages coming from bots rather than people. Many of those are AI crawlers reading your page to answer somebody's question somewhere else. If your content only exists after JavaScript runs, or only makes sense visually, it's not going to make it into that answer.

It overlaps with SEO and it's not the same job. SEO has traditionally been about ranking in a list of links. This is about being legible to something that reads your page and then writes a summary of it. People call it AEO or GEO, answer engine or generative engine optimization. Mostly it comes down to writing clearly and marking up honestly, which is the part that never changed.

Less than the internet will tell you. Google retired FAQ rich results on May 7, 2026, so the dropdowns are gone for everyone. The FAQPage markup is still valid and there's no reason to strip it out, but there's no confirmed evidence that adding it earns you citations from ChatGPT or Perplexity or AI Overviews. What helps is the visible answer being clear, complete, and sitting in the HTML where a crawler can find it. The markup describes that. It doesn't substitute for it.

Google Lighthouse is the honest baseline: performance, accessibility, best practices, SEO, four scores, hard to argue with. It doesn't measure everything that matters, and a perfect score doesn't mean the content is good. It does mean nothing structural is broken, which is a real thing to be able to say.

Maybe not. Design systems tend to be good at components and less good at what happens across pages, which is where headings drift, labels go missing, and the content model stops matching what people are actually publishing. If that sounds familiar, there's work here. If it doesn't, there might not be.

Legal will tell you the standard. They won't tell you where your product fails it, and they usually get involved after somebody complains. The overlap with machine readability is the useful part: the work you'd do for a screen reader is largely the same work that makes your content usable to an AI agent, so it stops being a cost center.

Some do. It's worth checking rather than assuming, because it degrades quietly. A quick audit will tell you in a day or two, and if everything's fine you've spent very little to know that.