Agent Teams

[MShipleyDP]

I am curious what your thoughts are on the potential of utilizing the new agent teams feature that they recently sent out.

[avifenesh]

@MShipleyDP
I need to learn it first. I'm adding what I genuinely use. Normally when I notice that I repeat some flows, semi-flows, tasks etc.
I haven't used it yet, I believe that by the end of the weekend ill have some thoughts, and then ill start my research about best practices.

I guess one problem can be the amount of tokens it will require, and while I can see how it is helpful for initiating a new project from scratch, I'm not sure I see how stable it is for projects that are brownfield prod projects.
But I love trying new features, it will be fun :) then I will have more info

What is your take?

[avifenesh]

@MShipleyDP After many back and forth - if you want to fast scaffold something, teams are cool, but for real code that needs to get into production i prefer the orchestrator with subagent and explicit workflows.
If I want to, which i do, to perform many tasks in parallel, opening few terminals with /next-task end up with results which has clear better quality.

In addition, teams share the context with each other, where in most cases it is wasteful.
I don't need an x feature to know the status of y. I have my checkpoints files, they can get a compact picture of what happens generally and this is generally enough.

I do use teams when I'm not in a structured flow and i want fast exploration of a large repo.
Didn't find a good case for it otherwise.

Did you have a different experience?