Blog
Essays and observations on managing science, the AI paradigm shift, research infrastructure, and what it means to create the conditions for discovery.
The Product Was the Headline. The Lesson Was Somewhere Else.
Two take-homes from Anthropic's AI for Science briefing and the Claude Science launch: the value is the integration and the harness, not the model, and acceleration only compounds when a laboratory checks it against reality.
AI for Science Needs Research Infrastructure
AI for science is two different things: models of nature, and systems that do research. Both are limited by data, and the data they most lack is the tacit, multimodal record of how science is actually done. That record is made in the laboratory, which puts research infrastructure on the critical path, and gives Europe a real opening.
Mapping the Self-Driving Lab Landscape
Six months of tracking who is actually building self-driving labs, distilled into a hand-curated interactive map. 74 entries across national programmes, academic groups, commercial vendors, and lab-OS platforms. Why I built it, and what it reveals about the field.
Building Plaudio
Notes on shipping a small open-source tool that labels meeting speakers by voice instead of by cluster. Why pyannote's default fails on similar voices, what voice-bank-first sliding match does instead, and three observations from building for myself first.
What Does It Mean to Manage Science?
Science management is not administration. It is not grant chasing. It is creating the conditions for discovery that would not otherwise exist. A first attempt at articulating the practice.