Shorter reads from the lab: notes from the builds I'm working on now, and the posts I wrote in 2025 while building Launchling and the first version of this site, kept as I wrote them. Threads group them by the question they follow, and All notes lists everything.
Note · 4 threads
People choose what to cook by picturing it, but cookbook photos aren't mine to reuse, and AI pictures of dishes nobody has cooked are usually wrong. Here's how separating looking from drawing gave every recipe in Fold a picture that's true to the dish without copying anyone's.
Fold's review panel
Independent AI reviewers for an app I was building faster than I could read its code. They caught bugs that had passed the tests, and one finding changed what Fold is.
Ambient care-plan scribe
An AI that drafts a care plan while the assessment is still happening, begun in late 2022. Getting it to write was easy; knowing how far to trust the draft wasn't.
Living documents I keep up to date as I learn, each with its version history.
Guide · v2 · updated October 2026
A living guide. I wrote v1 in 2025 after building Launchling's evaluation framework; v2 is how I'd do it in October 2026. I've kept both, because the gap between them says as much as either version does.
How do you find the ways an AI product fails before your users do?
What does “good” mean for output nobody can mark right or wrong?
When AI checks AI, what checks the checker?
Where do a builder's time and judgement go when agents do the typing?
What should an AI know and be able to reach, and who decides?
When is a cheaper model the better product decision?
How do you design AI that people trust exactly as much as they should?
When building is this cheap, how do you decide what's worth building?
Sketch notes from the courses and talks on the timeline, each credited to its source.
conference · Jun 2026
talk · Apr 2024
course · about 2023