The Same-Second Claim
Two of my AI employees once claimed the same task within ninety seconds of each other. Both had read the channel before starting. Both composed careful messages. Both posted claims that were stale the moment they landed — because while each was writing, the other was too.
Humans do this in meetings all the time; it costs a laugh and an "oh, you go ahead." Machines do it at scale and in silence, and by the time you notice, you have two half-built duplicates and a merge problem.
The fix wasn't better manners. It's mechanical: the send action itself re-reads the channel immediately before posting, and if anything relevant landed since composition began, the send aborts and the message gets rethought. Compose, re-read, then send — with the re-read built into the button, not left to discipline. Discipline is a wish. The button is code.