Are you delegating to AI, or handing it the decision?

Psychology has a name for what happens when choosing feels optional.

Two PMs, same Monday morning. Same sponsor email: "We need an updated plan. This is falling behind." Same AI tool open in the next tab.

PM A pastes the current plan into the chat. "Rewrite this so we're back on track." Two minutes later, a clean revised schedule comes back, dates shifted, milestones renumbered, nothing obviously wrong. She scans it, likes what she sees, sends it to the sponsor before lunch.

PM B pastes in the same plan. Before asking for a rewrite, he asks a different question: "If you were the sponsor, what would you push back on here?" The answer isn't clean. It surfaces a dependency on a vendor deliverable nobody has confirmed, and a milestone that only looks on-track because two tasks were quietly descoped last month. He spends twenty more minutes than PM A did. What he sends the sponsor is a plan he can defend in the room, not just one that reads well on a screen.

Same tool. Same morning. Same stakes. PM A handed the decision over. PM B borrowed the thinking and kept the decision.

That relief PM A felt, watching a clean plan appear with no more effort than a sentence, has a name. Psychologists have been studying it since long before anyone typed a prompt.

The Thing You're Not Supposed to Give Away

Psychologist Albert Bandura spent much of his career defining what he called human agency: the capacity to make intentional choices and exert control over your own actions. He built it on four properties — acting with purpose (intentionality), thinking ahead about consequences (forethought), regulating your own conduct as you go (self-reactiveness), and reflecting on whether it's working (self-reflectiveness). His conclusion, stated flatly: "the capacity to exercise control over the nature and quality of one's life is the essence of humanness."

Bandura also named the legitimate exception, and the term is the most useful thing in this entire essay: proxy agency — relying on someone else to act on your behalf. The authorship stays yours; only the execution moves. Hiring a lawyer to negotiate a contract. Asking a colleague to run a meeting you can't attend. That's how any complex life actually gets lived.

The distinction matters because giving up control was never supposed to be easy. Neuroscientists Lauren Leotti, Sheena Iyengar, and Kevin Ochsner reviewed decades of animal, clinical, and brain-imaging research and concluded that the perception of control is a biological imperative, wired into how organisms survive, as close to the operating system as instinct gets.

Which is what makes PM A's two-minute relief worth noticing. She didn't experience it as giving anything up. She experienced it as being efficient.

What Actually Happens in the Handoff

There's a well-documented mechanism for what PM A did, and it has nothing to do with AI specifically. Cognitive scientists call it offloading: using an external tool to reduce the mental work a task requires. Lists, calculators, GPS, and now AI drafts are all offloading, and offloading is often exactly the right move. The research is also clear that offloading is habit-forming: the more often a tool carries the thinking, the less practiced the underlying judgment becomes.

Pair that with what researchers studying automation call automation bias: treating a tool's output as a shortcut for vigilant review instead of a starting point for it. In cockpits and hospitals, this produces two failure types: missing a problem the system didn't flag, and following a bad recommendation because it arrived pre-packaged and confident. A revised project plan that reads cleanly invites exactly this. Nothing about it screams "check me."

This is the same failure, wearing a PM costume: reaching for the Hammer before doing the work of the Mirror and the Telescope, before asking what you're assuming or what you're not seeing from where you sit. Skip both, and AI dresses unexamined assumptions in a cleaner font. The thinking underneath the plan is exactly as untested as it was before you opened the chat.

A risk register drafted entirely by AI and shipped without a single line the PM argued with is a plan wearing the shape of judgment it never received.

Delegating on Purpose

PM B's twenty extra minutes bought something specific: the difference between proxy agency and something that only resembles it.

Here's the test worth carrying into your next AI-assisted decision: before you delegate it, can you say what you'd do differently if the AI disagreed with you? If you can answer that, you're still the author, the same way you're still the author of a contract your lawyer negotiated on your behalf. If you can't, the decision stopped being yours somewhere between the prompt and the send button.

The stakes for building this habit now aren't hypothetical. In a 2023 Pew Research and Elon University report canvassing 540 technology experts, 56 percent predicted that by 2035, AI systems will not be designed to make it easy for humans to stay in control of most tech-assisted decisions. That's expert forecasting, not settled fact, and the remaining experts split the other way. But it's a forecast about default settings, and default settings are exactly what people stop questioning once they've stopped asking whether they still agree with the decision being made.

AI was never going to take your agency by force. It doesn't need to. Every deferral is voluntary, handed over the same way people have always handed it over: because deciding takes effort, and for a moment, not deciding feels like the same thing as moving fast.

What's one decision this week you let AI make that you never actually decided to delegate?

Get Intentional,
Paul

Let's Talk! I want to hear where your struggles are. This is a conversation. Not a pitch.