Do you use multiple agents for software development?
I generally develop my projects with one agent. I know they are not big, but it seems enough for me. Have my own methods for keeping context and to make agent reliable. Is it really necessary using multiple agents?
1 コメント
I vibe coded bukowski [1] to circumvent the Claude Code scrollback bug... then added support for codex and gemini in multiple panes... then an MCP server for them to chat with each other. I usually ask them to comment on the plan and brainstorm before a Claude instance commits to implementation, then the others review. Sometimes even two Claudes come to different conclusions or refine each others ideas, Codex is great at finding edge cases or non obvious bugs.
[1] https://github.com/vmitro/bukowski
[1] https://github.com/vmitro/bukowski