Theoretical physicist Matthew Schwartz released BootLoops 1.0 on October 1, 2026, an open-source toolkit that automates exact calculations across quantitative science. Built with Claude Fable 5, the system reproduced one of Schwartz's own papers in 20 minutes and has already solved 15 previously unsolved elliptic Feynman integrals - a concrete benchmark that signals a shift in how AI can accelerate research workflows.
Schwartz began the project after Anthropic released Claude Fable 5 in Summer 2026, following an earlier experiment with Claude Opus 4.5 in December 2025. The toolkit runs on Google Cloud VMs, coordinating Claude Code sessions while managing safeguard triggers that might otherwise interrupt long-running scientific computations.
What BootLoops actually does
BootLoops 1.0 automates exact calculations across 18 fields, including physics, ecology, and genetics. In three months, the system validated 730,000 human exomes and produced 36 manuscripts with 19 coauthors - output that typically requires far larger teams and much longer timelines. The toolkit is released under the MIT License, with prose and figures under CC BY 4.0.
Anthropic funded the project, but Schwartz maintains it independently. The code has been validated on Linux x8664 and arm64 architectures. Schwartz is clear about the boundaries: the toolkit is not officially supported by Anthropic and is intended for research, not clinical or regulatory use.
A growing pattern in AI-assisted research
Schwartz's results fit a broader trend of domain experts using large language models not as replacements, but as force multipliers for technical work. The 20-minute paper reproduction isn't a parlor trick - it demonstrates that structured scientific reasoning can be accelerated when the right scaffolding exists between researcher and model.
The choice of MIT License matters here. By keeping the toolkit open, Schwartz allows other research groups to replicate and extend the workflow without licensing friction. The validated hardware targets also mean teams can deploy on standard cloud infrastructure without specialized setups.
Why this matters for science and research professionals
BootLoops represents a practical template for integrating AI into rigorous quantitative work. Rather than chasing general-purpose chatbots, research teams can examine this toolkit's architecture to understand how domain-specific automation actually functions - from managing model context windows to handling the verification steps that make outputs trustworthy. For professionals building or managing research pipelines, the open-source code provides a reference implementation that can be adapted to fields beyond the 18 Schwartz has already demonstrated. Those looking to build similar capabilities can explore AI for Scientists Courses or AI Agent Courses to develop the technical foundations this kind of work demands.
Your membership also unlocks:
