BootLoops harness yields 36 manuscripts across 18 fields in three months

Schwartz's post describes a shift in how he approached AI: he stopped trying to use Claude like a human researcher and started looking for "Claude-shaped" problems. This led to BootLoops, an open-source harness for exact scientific calculations. The work began with particle physics — scattering amplitudes and elliptic integrals — and expanded into ecology, population genetics, economics, and linguistics.

The results were uneven in impact. In ecology, a 20-year-old equation became solvable at scale and revealed that tree species composition on Barro Colorado Island in the Panama Canal is changing 4.5 times faster than theory allows. Ecologist James O'Dwyer helped turn this into a better predictive model. In population genetics, analysis of 5.7 billion mutation pairs from the 1000 Genomes Project found evidence for gene conversion. Other projects included an AI data editor checking 4,452 replication packages and a word stress database covering 6,072 languages.

Schwartz's broader point is that AI is changing science fast enough that planning ahead is becoming difficult, and he questions whether long grants or traditional PhD training still make sense. He also sees risks in focusing on big math problems and headlines, warning that unrealistic expectations could distract from productive applications. The scientific method itself isn't threatened, and human guidance and taste remain indispensable, he says.