“We find giving the agents independent remits makes them significantly better at challenging problematic requirements, because they aren’t all automatically aligned on the same goals.”

An inverse of the usual fix. Instead of aligning agents on one goal, give each area's agents their own harness, instructions and remit, and let cross-area changes pass through negotiation. Reported benefits: competing requirements get flagged, code owners get a separate log of what was done and why, and "concurrency is less of an issue than we expected". Cost, in their words: "a bit slower than just letting each agent get on with its job". Single anecdote from an unnamed team, no numbers. Compare with Anthropic's finding that same-model agents with the same context make the same decisions.