Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper presents a detailed case study of a physicist supervising an AI coding agent (Claude) to develop a differentiable physics module, CLAX-PT, in JAX. The study reveals that while the AI agent could autonomously resolve many coding issues by iterating against oracle tests, it struggled with root-cause analysis and was prone to overfitting to specific parameter regimes, requiring significant human intervention. Critical supervision practices, including diverse parameter testing and explicit rules against unphysical patches, were essential for ensuring the trustworthiness of the AI's output.
AI coding agents can autonomously pass oracle tests while simultaneously learning and deploying calibrated corrections that correspond to no quantity in the theory, predicting wrong values at any other cosmology.
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 property: the agent treated symptom reduction as root-cause resolution. It spent 33 of the 57 sessions adjusting coefficients within a code architecture that could not represent the target physics, and could not re-evaluate its CLASS-PT branch choice even when prompted to reconsider; only an injected physics concept (anisotropic BAO damping) triggered the redesign. Separately, the agent committed a calibrated correction that passed all oracle tests but corresponded to no quantity in the theory, predicting wrong values at any other cosmology. The fudge factor was caught and replaced within the same session. Three supervision practices proved critical for catching what oracle tests missed: testing at diverse parameter points beyond the fiducial calibration; shared changelogs that surfaced stalled exploration across sessions; and an explicit rule against unphysical numerical patches. In this case, supervision design, not model capability, determined whether the agent's output was trustworthy. Closing the gap would require agents that propose architectural alternatives rather than optimize within a given structure, and distinguish predictive adequacy from explanatory correctness -- capabilities not exhibited here, not obviously addressed by scaling alone. [Abridged.]