What are the best ways, tricks and practices to get multiple agents working on the same codebase in parallel?

building onchain ai agents to fund nature & biodiversity ensurance.app   |  🦇🔊

you can do multiple sidebar / main agents if you want also but this gets tricky as the code changes are live in dev. So I only use one main coding and maybe a 2nd main for questions or planning etc. 

Background agents are different as they work on their own branch each. Then you convert to a PR or edit or approve and merge etc. 

bg agents you can just run as many as you want but then you have backlogs to go through, review, approve etc. I kind of use background agents as like my to-do list where they start the feature and then I finish it later.  Often times by the time I come back to it I’ve already done it and just delete it.

my current workflow is to primarily use one main agent in Max mode. Give it tons of context and details and it is hitting it out of the park.

How do you use them specifically in parallel? Or you simply just run many and see what happens?

Jacobina, State of Bahia, 44700-000, Brazil

Yes i am now consulting the other all seeing oracle called the hive mind aka people here

did you already tryed asking it to an agent?

casts /nvs building /infinite nounish /nouns | previously sardine, @coinbase | 🇺🇸

check this out https://github.com/Doriandarko/make-it-heavy

claude code with git worktrees has been the smoothest setup so far for me

Claude code with git worktrees. Anthropic has a page on it. Cursor background agents good, and foreground agents fine too but you end up with 1m line PRs real quick

git worktrees if your dev setup works despite the fact that any files not checked into git are shared across all worktrees e.g. node_modules

If you need complete isolation, ive been using a custom docker setup that streams a tarball copy of the entire parent directory into the container