The Founding.

I run an eighteen-person design practice with four managers. I also run a constellation of AI agents that handles research, editing, critique, and daily triage. The site you’re reading was supposed to be a portfolio. Today I changed my mind about what it should be.

The research phase that led here ran in a single day. The Librarian mapped the territory: thirteen voices, checked against what they’d actually published in 2026. The Researcher read demand data: eight questions design leaders are asking right now, scored against the evidence I already have. A third pass tore down the sites of Simon Willison, Maggie Appleton, Linus Lee, and Addy Osmani to understand what the builder-in-public genre actually does and whether it transfers to someone running a team rather than building tools alone.

The questions

What do I do with the time AI gives back?

25 / 25

How do I stay credible when AI makes me feel early-career again?

20 / 25

How do I maintain craft standards when AI lowers the average bar?

16 / 25

The top three of eight questions design leaders are asking, scored demand × lived evidence out of 25. The highest-scoring question already had a finished essay against it. Demand brief, 7 Jul 2026.

All three sessions pointed to the same gap in what we’d reviewed. We found commentators describing where design teams should go, researchers surveying what teams were doing, and individual practitioners documenting their own changed workflows. We didn’t find a creative director at a design consultancy publishing an ongoing account of changing their team’s practice, with records of what held and what broke.

I’ve been sitting on that material for weeks. Essays drafted. Cases built. A production system running daily. And a site that presents it as finished work, as credential. That framing was defensible. Hiring decision-makers need something to inspect, and finished cases give them that. The finished cases weren’t showing the practice as it changed. The interesting work is the search itself.

So the site shifts. The credential layer stays: cases, verified metrics, plates that demonstrate the hands still work. But the expedition log leads. The typeset record of a running practice, published as it runs.

The decision has a cost I priced in before making it. A log that stops reads as an abandoned practice, worse than no log. And every note requires an editorial gate: publish the reasoning and the system, and keep the client and the specifics out. That gate has no expiration date. I accepted it because the bought side of the trade is structural: the site, the OS, and the positioning stop being three separate projects. Every week the practice runs, the site gets truer.

The trade

Cost

A cadence obligation with a sharper failure mode: a stopped log reads as an abandoned practice. A per-piece editorial gate, no expiration date.

Bought

The site, the OS, and the positioning stop being three projects. The differentiator compounds in public. The empty seat, claimed while it’s still empty.

The decision’s accounting, written before the decision was made. Decision record, 7 Jul 2026.

One thing corrected itself mid-process. A first draft of this system overengineered the cadence: weekly notes, aggressive publication targets, the kind of velocity that sounds rigorous on paper and produces thin work in practice. I pushed back. A monthly note that says something true beats a weekly note that says something thin. The log serves the practice. We settled on monthly notes, each carrying records the reader could inspect.

This note is the first instance of that. The log begins by logging its own founding: the research that found the whitespace, the decision that chose to fill it, and the correction that kept the system from becoming another thing that needs managing. That’s the practice deciding: research, pattern recognition and a redirect when the evidence says so.

The seat was empty. I sat down.

Researched by the Librarian and the Researcher. Drafted with the Editor. Decided and passed by my eye.