Physics Is All You Need? A Case Study in Physicist-Supervised AI Development of Scientific Software
arXiv:2605.30353v1 Announce Type: new Abstract: Are AI agents tools, co-authors, or researchers? We present a quantified case study ($N=1$): a physicist supervising an AI coding agent (Claude Code, Sonnet and Opus models) over 12 work days and 57 sessions to build CLAX-PT, a differentiable one-loop perturbation theory module in JAX. We documented and classified 15 supervision events by intervention level. The agent resolved ten autonomously by iterating against oracle tests. Two more by the physicist's domain knowledge. The three it could not -- all evaded oracle detection -- share a common pr
This case study details recent practical progress in AI agents directly assisting scientific software development, demonstrating increasing capabilities for autonomous work and human-AI collaboration.
It illustrates the tangible application of AI agents in complex, specialized fields like physics, pointing towards a future where AI significantly augments, if not replaces, parts of white-collar workflows.
The interaction model between human experts and AI agents is evolving from simple tool usage to more collaborative, even supervisory, roles for humans over increasingly capable AI systems.
- · AI agent developers
- · Scientific software engineers
- · Physics research
- · Entry-level software development
- · Manual coding of routine scientific modules
Physicists and other scientists will become more efficient in developing specialized software and models.
The pace of scientific discovery and software development in specific domains will accelerate due to AI assistance.
This could lead to a redefinition of scientific roles, with humans focusing more on high-level problem-solving and AI handling implementation details.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI