AI agents (LLMs) are very good at automating the less-fun parts of programming. They're also very good at working in a loop: try something, verify if it worked, learn and try again. We'll talk about how to harness LLM agents to fix performance problems, even in Ruby itself (and even if you don't understand C!).

We'll show how to use a combination of LLMs, simple scripts, skills and MCPs to create reproducible benchmarks. These tools will combine to create an "agent-native" Ruby performance improving workbench. We'll also explore and discuss the limitations of these tools, including what kinds of problems they tend to be bad at solving.