NEWESTBLOG
The screen-building job is dying. UX isn't.
Chat and generated interfaces are hollowing out the work of hand-building screens. The same evidence shows where UX work moves next: into the parts agents assemble from, the rules they follow, and the judgment of what they produce.
Blog
- The screen-building job is dying. UX isn't.
- I tested 11 Mandarin speech models. The most accurate can hide the mistakes
- What I delegated to AI agents in a production release — and the three things I kept
- 13 launches, 0 bugs in the release, 1 switch: a release runner you can hand to an agent
Podcasts
- RLAIF vs. RLHF - Scaling Reinforcement Learning with AI Feedback
- Frontier governance frameworks and independent assurance
- Building a safety case
- Ecological validity, selection effects, and uncertain base rates
Projects
- fidx
local hybrid semantic search that beats QMD on recall and precision at ~300–1000× lower latency.
- token-count-compare
a side-by-side tokenizer comparison across frontier models.
- token-usage-analyzer
finds what's burning your token budget in AI coding CLIs.