Agentic Coding For Scientists

I’ve been fielding lots of questions from scientists who are interested in using coding agents like Claude Code, so I thought I’d make a blog post. That became four blog posts!

This series is written from the perspective of someone who works with public data and writes open source software. If you work with sensitive data, please take appropriate precautions. Agents are random and could act maliciously. Caveat emptor.

1. Agentic Coding from First Principles

October 30, 2025

Understanding how coding agents actually work leads directly to strategies for using them effectively. This post explores the fundamentals and shows how good software engineering practices become even more important when working with agents.

2. Agentic Git Flow

November 1, 2025

Git Flow is an industry-standard development workflow. This post describes how to adapt it for agentic coding, with detailed planning documents, iterative review cycles, and multiple quality gates to ensure that code comes out exactly as you want.

3. Writing Scientific Code Using Agents

November 4, 2025

Scientific programming has different requirements than typical software. We want code that fails fast, prioritizes correctness over robustness, and treats backwards compatibility as clutter. This post covers the special considerations for using agents in scientific work.

4. The Human Experience of Coding with an Agent

November 7, 2025

Beyond the technical details, how should we work with an AI agent day-to-day? This post explores the human and philosophical dimensions: flow state, balancing emotions when interacting with agents, and the bigger ethical picture.