Suggestion: Multi Agent (Supervisor/Reviewer agent) about aider HOT 2 OPEN

Thanks for trying aider and filing this issue.

I have experimented with this type of approach and not seen quantiative benefits.

Aider is currently the top scoring agent on SWE Bench. So there's some reason to trust that it's current strategy is solid.

https://aider.chat/2024/06/02/main-swe-bench.html

from aider.

@paul-gauthier I can appreciate your perspective, I just want to add some supporting evidence for the case of Agents outperforming raw model outputs. I think as a leader in the code assistant space, with all the benchmarks established it would be extremely valuable to the community if aider was to expand the benchmarks to include a leader board for more complex agent workflows. I imagine swe-agent, open-devin, and MoA would be great initial candidates for this benchmark.

https://github.com/togethercomputer/MoA

We are looking to do something like this internally currently, and I'll happily share what we get working in due course if no one else beats us to it but, this wouldn't be a substitute from a section on aiders blog that reports the results

from aider.