Part of what drove me to having as many lanes is that I was going insane waiting for the model to do its thing, and at some point I could not make the interactions meaningfully faster.
I've never had that much trouble going from one thing to the next, and diffs are usually small enough that I can review them easily. If something feels like it is getting out of hand I can always pause and actually review on GH.
That being said, this is not really about speed, rather releasing smaller chunks sooner, there is some speedup, but it is mostly in that I can get to feedback faster and I can do more things at the same time (at maybe the cost of all going a tad slower?)
I hope this makes sense
Kartik N V J K
AI Developer | Making AI reliable, trustworthy & accessible to everyone | Active community contributor
Running smaller steps across several agents is interesting, and the wrong-unit point resonates: the review unit should be the diff, not the task. I score each diff on its own so a six-agent workflow does not become six places to miss a bad change. How do you keep the overhead of reviewing that many small steps from eating the speed you gained?