Field Note
Onboard It Like a Hire. Own It Like a Commander.
I’ve spent months telling you to manage your AI like a junior employee. Plan for it, supervise it, onboard it. I stand by every word.
But there’s one line the metaphor can’t cross — and crossing it is how good operators get burned.
Call an agent an “employee” and something measurable happens: in HBR’s study of 1,200 managers, framing it that way made them catch 18% fewer of its errors, and quietly shift the blame to the model when it failed. The accountability didn’t actually move. It never does.
You can delegate the work. You can’t delegate ownership of the outcome.
I learned that vocabulary in the Marine Corps, not on an org chart. A commander doesn’t hand responsibility for an operation to the operations officer who planned it. The OPSO designs the op; the commander owns what happens when it meets the enemy. Delegation moves the task down. Responsibility stays at the top. Always.
An agent is no different. It can draft the package, run the analysis, build the thing in minutes — real work, genuinely done. But the moment it ships, the deliverable is yours. Not the model’s.
Here’s the trap, and I’ve walked straight into it: the better these models get, the easier it is to stop reading what they hand you. You skim. You approve. You “trust the team.” And the one time it drifts — confidently, past the ragged edge of what it actually knows — that’s yours too. You just stopped watching for it.
So onboard your AI like a junior hire: a role, scoped access, guardrails, a record. That half of the metaphor holds.
But own it like a commander. The judgment stays yours. The outcome stays yours. That half never transfers.
Before your next agent ships something that matters, ask the question rank has always forced: if this goes wrong, whose name is on it? If the honest answer is “the AI’s,” you’ve already lost the thread.
Onboard it like a hire. Own it like a commander. The first half you can delegate. The second half never leaves you.
Want your own AI workflow checked for where ownership actually sits — or a look at the plain-language readiness checklist I’m building? Start a conversation.
Source: HBR, “Research: Why You Shouldn’t Treat AI Agents Like Employees” (Kropp, Bedard, Wiles et al., BCG + BU Questrom; May 2026) — framing agents as employees cut managers’ error-catching by 18% and pushed perceived accountability toward the model, though it never actually leaves the humans who deployed it. The command-responsibility framing is mine.