An agent that never sleeps is the whole promise of this wave of AI: something that watches your inbox, your repo, your calendar, and acts while you’re asleep. The problem isn’t whether it works. The problem is the bill.
The meter never stops
A cloud agent looping every few seconds, calling a frontier model each time, adds up fast. Run the numbers on a genuinely always-on loop and you land somewhere near $50/day — call it $1,500/month for one agent. Multiply by a team, or by the dozen small agents you’d actually want, and “always on” quietly becomes “on during business hours, because finance said so.”
What local changes
Mint a small model, ground it in the library that actually matters for the task, and run it on hardware you already own. The marginal cost of another inference is electricity. A 24-hour loop lands near $0.03/day — three cents.
That’s not a 2× saving. It’s three orders of magnitude, and it changes what you’re willing to leave running. Cheap enough to be always-on is a different product than expensive enough to ration.
Own the machine, stop paying the meter.
The catch, stated honestly
A minted local model is not a frontier model. It’s narrower on purpose — it knows your library cold and abstains outside it. For an always-on agent doing a bounded job, that’s the right trade. For open-ended reasoning across everything, it isn’t. werbos is built for the first case and honest about the second.