PSI Inc. — a Boston and San Francisco startup operating as Physical Superintelligence PBC — has released what it claims is the first open-source agentic AI system built specifically for physics research. The tool is branded "Get Physics Done" (GPD) and surfaced via a YouTube video that reached the Hacker News front page. That video is currently the primary public record of the announcement; it provides a product overview but limited technical detail.
GPD is pitched as an AI copilot built by physicists for physicists, capable of autonomously running physics-related reasoning and research workflows. Most deployed AI agents today target coding, productivity, or <a href="/news/2026-03-15-zirco-ai-launches-ai-employee-for-dental-front-desk-operations">customer service</a>. GPD takes aim at one of the most technically demanding scientific disciplines instead. PSI's stated mission is to industrialize scientific discovery at scale — what the company calls "a vertically integrated factory for physical superintelligence" — with a long-term goal of shifting physics from incremental progress to systematic, rapid breakthroughs.
The team draws contributors from Google, OpenAI, Harvard, MIT, Stanford, Oxford, Johns Hopkins, Cambridge, the Institute for Advanced Study, and the Perimeter Institute. PSI has not named a lead founder or principal researcher in this announcement, which makes the credential list difficult to verify independently. The company says it intends to commercialize transformative physics discoveries safely and for broad public benefit.
Going open-source is GPD's clearest differentiator from tools like Elicit or Semantic Scholar, which operate as closed platforms. A <a href="/news/2026-03-15-nova-self-hosted-personal-ai-dpo-fine-tuning-autonomous-self-improvement">public codebase</a> means physicists can inspect GPD's reasoning chains, fork it, and audit outputs — a meaningful distinction in a field where reproducibility is foundational. Open-sourcing also exposes the work to scrutiny from both the research and AI communities, which will stress-test PSI's ambitious framing faster than any proprietary rollout would. The next milestone to watch is whether PSI publishes technical documentation substantial enough to match the scale of what it's promising.