Not Every Team Needs a Dark Factory: How AI-Enabled Teams Actually Work

There’s a lot of excitement about loop engineering, dark factories, and agent swarms. Much of that work is impressive. It shouldn’t be every team’s goal.

SEP teams build software in regulated healthcare, aerospace, and pharmaceutical environments where the cost of being wrong is high. We’ve found that constraints like token budget, verifiability, complexity, predictability, and code health determine the best operating mode for a team. Some constraints sit outside a team’s control. The rest move only with deliberate investment, and that’s where to start. Teams that copy someone else’s mode without addressing their own constraints tend to see mixed results.

In this talk, we walked through the operating modes we’ve applied successfully on real teams, from individual agents running ad hoc on a laptop, to a shared harness committed in the repo, to orchestrated long-running agents. For each, we showed how the constraints made the call. The differences are what you verify, code or outcomes, and what humans manage, keystrokes or systems.