# Points of improvement

- Running benchmarks through GH. List in your PR which benchmakr you're trying to improve
- Comment: `/workflows/benchmarks <optional-test-name>`
- Don't forget to run other benchmarks that may be effected.
- Sometimes fixing one benchmark may break others. Test for regressions locally more strictly.
- Leverage the separation of agents to more easily test for regressions
- Focusing more on benchmarks to make them consistent.

---------

# Continue to do:

- If you have an intuition that you can get a PoC of in less than a few hours; it's a better and more efficient demonstration of your point than plain talking.
- Use ChatGPT for prompt tailoring
- Explore ChatGPT plugins for prompt optimization (ex: PerfectPrompt (?))
- Explore leaderboard agents which are very good in a specific area and see what they do for their prompting