Running multiple agents

After working through a full PRD and epics, sometimes you get started and notice by epic 3 that you should have done something differently. What should you do? It depends on the severity of the change. Sometimes a simple course correction will do. If it is a major shift, consider cloning the repo again and starting at an earlier point before the mistake. Out-of-order epics? Restart after architecture. Does it trace back further? That is okay — sometimes it makes sense to keep what is in progress going.

Start a new track by cloning the repo and beginning from a commit that makes sense. Create an entirely new branch and keep both efforts running in parallel.

My most important deliveries usually have three separate parallel efforts all going at the same time. You can run gap analysis between branches to cherry-pick the work you like most. This strategy scales further by having multiple agent operators developing the solution — each track gets its own operator, and you converge on the strongest output.

The key has always been the delivery. Don’t rabbit hole yourself into one track.