Paid to hand over the keys.

A workforce you have to operate is a tool that grew a staff.

That's the test, and it's worth running on everything currently sold as an AI workforce. Who runs it? If the answer is a specialist - a new job title, a certification track, a hire you make so the thing produces - then the operating hasn't gone anywhere. It moved from your team to a priesthood.

The tool era kept minting priesthoods. Whole professions exist because powerful software needed skilled humans composing it, feeding it, and gluing its outputs together - the GTM engineer, a person whose full-time craft is wiring go-to-market tools into workflows and keeping them running, is a real job with real skill. No knock on the people; the craft is genuine. But the existence of the profession is a measurement of the software. When a product creates a new profession to run it, the product has told you exactly where it ends.

When a product creates a profession to run it, the product has told you where it ends.

A real workforce absorbs the running. And when it does, something more interesting than replacement happens to the management around it. The classic manager's job was always two jobs wearing one title. The first is translation: turn the owner's intent into tasks, sequence the work, chase the status, roll it back up. The second is judgment: decide what's worth doing, approve what's risky, gauge what has earned more trust, answer for the result. The workforce takes the first job almost whole. It cannot take the second.

The manager's job was two jobs wearing one title. The workforce absorbed the translation. The judgment became a seat.

What's left is not a layer of the org chart. It's a seat. One person who sets the goals, approves the bets, decides how much rope has been earned, reads the results, and carries the owner the few calls only an owner can make. That seat deserves a name: the AI Workforce Engineer. AWE, if the era wants an acronym - and eras always do.

The name is chosen carefully. Software already celebrates a role called the forward-deployed engineer - the specialist a vendor embeds in your company to make the vendor's product land. Honest, skilled, well-paid work. But it faces the vendor. The AI Workforce Engineer faces the business. Nobody deployed them. They sit on your side of the table, and they answer for outcomes, not adoption.

And here is what separates this seat from every operating role that came before it: it's designed to shrink. Early on, the person who brought the workforce into the business usually holds it - setting the first goals, approving the first moves, watching the first results come back and checking them against what was promised. As the workforce earns trust action by action, the seat gets lighter and moves inside the business, to an owner or an ops lead, whoever is natural. Six months in, the review that took a morning takes twenty minutes. The seat keeps shrinking after the handoff, because the mark of governing a workforce well is that it needs less governing.

The seat is designed to shrink. That isn't a flaw in the role. It is the role.

Watch the money and the whole arrangement snaps into focus. The consulting era billed by the hour, and hourly billing has gravity: the rational move, for even the best people, is to stay needed. No villain required - the model votes for dependency. This seat pays the other way. The person who installs a workforce and governs it well is paid from what the workforce produces, not from time on a clock. Handing the seat over doesn't end the relationship, and it doesn't end the earning. It frees them to go start the next one.

The old model paid experts to stay needed. This one pays them to hand over the keys.

There's a note earlier in this series about handing the wheel to the workforce. This is the other handoff - the human one. The keys change hands twice: once into the seat, once out of it. The businesses getting the most from this era aren't the ones that hired the best operator for their AI. They're the ones where, seat by seat, nobody has to operate at all.

Dependency was a revenue model. Autonomy is a better one.

Paid to hand over the keys.