← Back to blog

AI as a Scientific Copilot

LLM agentsClaude Coderesearch workflowscientific computing

Claude Code has become my scientific copilot.

I’ve been an active user since it launched, and it has genuinely changed how I approach day-to-day research. What started as occasional code generation has turned into a full integration with my workflow — planning experiments, writing and debugging code, and iterating on statistical methods.

This semester, I gave two invited talks at the USC Biostatistics Department: one for students in the data analysis course and one for students and faculty in our seminar series. In both, I walked through how I actually use Claude Code in my research, not as a novelty, but as a core part of how I get work done.

The Talk

In my most recent presentation, I demonstrated how Claude Code enhances the research workflow end to end — from planning and code generation to verification and specification-guided agent evolution using CLAUDE.md.

Through live demos, I highlighted two key takeaways:

  1. Context is king. The more structured context you give the agent — project specs, coding conventions, known constraints — the better the output. This is where CLAUDE.md shines: it acts as a persistent specification that shapes agent behavior across sessions.

  2. Verification is essential. AI-assisted research only works if the results are reproducible and trustworthy. I showed how I build verification steps directly into the workflow, treating every agent-generated output as a draft that needs validation before it becomes part of the research record.

The full talk is available below:

Why This Matters for Research

As Claude Code gets smarter, I see the productivity gains compounding. But beyond raw speed, the bigger opportunity is in agentic scientific discovery — using AI agents not just to write code faster, but to explore hypotheses, run simulations, and surface patterns that would take a human researcher much longer to find.

Several recent preprints are exploring this direction, and I think we’re still in the early innings. The researchers who learn to work effectively with these tools now will have a real advantage as the capabilities continue to improve.

Get in Touch

If you’re a researcher interested in AI agents for accelerating science, I’d love to connect. Reach out on LinkedIn or GitHub.