Line Your Own Pockets Using AI Agents for Strategies Part 3
Mar 30, 2026
They explore using AI agents to run massive backtests overnight and automate strategy iteration. Practical setup tips cover sandboxed mini-PCs, CLI control of trading tools, and safe permissioning. They discuss test-driven development for strategies, using example trades as pass/fail checks, and when to build versus using mature tools. Scaling workflows and automating code reviews also come up.
AI Snips
Chapters
Transcript
Episode notes
Mini PC Sandbox Keeps Experiments Safe
- Michael runs agents on a separate mini-PC sandbox not signed into his main accounts so worst-case damage is contained.
- He uses a mini PC with 32GB RAM and Git backups so Claude can experiment without access to personal data.
Require Unit Tests Before Features
- Use test-driven development and unit tests as guardrails: write failing tests first so the agent knows when a change succeeded.
- Dave May keeps unit-test rules in his main claude.md so Claude repeatedly runs and validates code changes.
Anchor Backtests With Example Trades
- Add example trades as unit-test anchors so backtests must include known trades or the run counts as a failure.
- Dave suggests specifying a concrete trade (e.g., Nvidia on a date) so Claude can verify inclusion and iterate if missing.