An Illustrated Guide to AI Agents

Chapter 10. Code Agents and Code LLMs

A Note for Early Release Readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 10th chapter of the final book. Please note that the GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Coding agents are some of the most potent types of AI agent in terms of the size of their likely impact. Code generation became one of the earliest applications of LLMs once they hit the scale of GPT-3 in 2020. As we saw in Chapter 3, code generation was also a major component for reasoning problems that reasoning LLMs tackle. Code is a key investment area for LLMs for two reasons. First, an enormous range of knowledge work, from software engineering to research, data analysis, and visualization, can be expressed as code. Second, and more useful for training, code can often be verified automatically: you can run it, or test it, and get a clear signal of whether it worked. That verifiability is what makes coding more tractable than open-ended tasks where there is no objective check.

This chapter walks through that world. We’ll meet the people who use and build code agents, see what makes software engineering agents distinct, build one ourselves, and finish ...