Andrea Belk Olson: Someone has to check the machine

The business case for adopting AI seems, on the surface, like a tidy equation. The tool costs this much, it saves that many hours, and the difference is your return. It's a persuasive argument. But it's also missing a number, one that doesn't surface until months later.

That is the cost of being responsible for whatever the machine produces.

The productivity gain is real. So is the new labor.

AI genuinely makes work faster. The first draft writes itself, the analysis runs in seconds. None of that is hype. But speed at the point of creation manufactures work at the point of approval. Every output produced faster now has to be checked — because a confident, fluent, completely wrong answer looks exactly like a correct one until someone with judgment examines it.

That examining doesn't happen for free, and it doesn't happen everywhere. The person who verifies the output, catches the error before it reaches a customer, decides whether this particular use of the tool is even appropriate, and coaches two employees who used it in opposite ways — that's usually a manager. The productivity landed on the team, but the accountability landed on them.

Load isn't what burns people out. Powerlessness is.

We assume people break under sheer volume of work. Yet the behavioral evidence shows what wears people down isn't how much they carry, it's carrying responsibility for outcomes they don't control. A manager who answers for AI-assisted work but had no say in which tool was adopted, no rule for when its output can be trusted, and no reduction in their existing workload is standing in exactly that spot. They own the consequences of a system they didn't design and can't fully predict.

That's attrition waiting to happen. It claims people who were never overwhelmed by the work itself, but by being left to mitigate every gray area by themselves.

Replace improvisation with a rule

The fix isn't to slow adoption or set up another training session. It's to decide on what this tool is actually for, and where its output can never go unchecked. The kind of work that can be done by the tool without review versus the kind that never leaves without a human signing off. This call belongs to the manager; that one doesn't.

The moment those lines exist, the manager stops being the lone arbiter of every judgment call and starts reasoning from a clear standard. That's the value of any real decision framework, converting a thousand improvisations into one agreed-upon rule. “Use AI” is not a rule.

You can keep treating AI adoption as a line item that only adds value. Or you can name the line item it also creates — the labor of making sure the machine was right, and decide, deliberately, who pays it and how. One of those approaches keeps your best managers. The other loses them, while the P&L still insists you're winning.