The Work That Cannot Be Delegated
Resistance is where understanding forms
Imagine you run an entire discovery sprint through an AI agent. The agent drafts the stakeholder outreach emails, synthesizes the responses, clusters the themes, writes the opportunity framing, and produces a prioritized recommendation. The output looks like the work. It has the right shape, the right vocabulary, the right level of detail. It is also completely wrong, in a way you won’t discover for three months, because you were absent from every moment of actual contact with the problem.
This is the failure mode the quote points at: you can outsource your thinking but you cannot outsource your understanding.
Most people read this as a reassurance about human irreplaceability. That is the wrong reading. It is a description of a cognitive mechanism, and it matters because agentic AI makes the failure newly accessible to anyone with an API key.
Thinking, as an operation, is separable from the thinker in ways that understanding is not. Jean Piaget, the Swiss developmental psychologist who spent fifty years studying how children construct knowledge, called the relevant process equilibration. It is what happens when a mental schema meets something it cannot process, destabilizes under the pressure, and reorganizes around the new information. That reorganization is understanding. The agent does not reorganize. It has no prior schema to destabilize. It produces outputs that look like understanding without any of the internal restructuring that would make those outputs load-bearing in a human mind.
The philosophical attack on this distinction is worth taking seriously. If understanding is just consolidated thinking, and thinking is pattern completion over a representational space, then the difference between what the agent does and what you do is a matter of degree and substrate, not kind. Martin Heidegger, the German philosopher who spent his career examining what it actually means to know and be in the world, would say this framing misses what is at stake. Understanding, in his account, is not a cognitive output you produce and then possess. It is a mode of being-in-the-world: the way things show up for you differently once you have genuinely encountered them. To understand a user problem is to have the complaint ticket feel different in your hands, to notice a pattern in a support call that you would have ignored six months ago. It is not propositional. The agent can represent the proposition. It cannot have its world change.
Al-Ghazali, the twelfth-century theologian and philosopher whose work on the limits of rationalism shook the Islamic scholarly world and later shaped the thinking of Aquinas, drew a sharp line between ‘ilm al-yaqin and ‘ayn al-yaqin: knowledge by description versus knowledge by direct encounter. He used the example of fire. You can read every account of fire ever written and still not know fire the way someone who has been burned knows it. The agent has read every account. It has not been burned.
In practice, the failure looks like this. You use the agent on the hard parts: the synthesis under ambiguity, the judgment calls about what matters, the framing of a recommendation. Those are exactly the moments where your mental model was supposed to meet friction. By delegating them, you preserve your existing model intact. Months later, when the model collides with reality, you have no mechanism for the collision, because you were never in the room where the friction lived.
The distinction survives the skeptic’s pressure because it is not a claim about consciousness. It is a claim about professional consequence. Understanding is what makes your next decision non-random. If you hand off the cognitive work that would have built it, your decisions revert to whatever priors you came in with. The agent wrote the strategy. You still own the priors.
This does not mean delegation is the problem. It means the question is which cognitive load to keep. The work worth keeping is not the tedious work. It is the work that would have changed your mind. The moments where the schema was supposed to destabilize. Delegate the research. Delegate the formatting. Delegate the first draft. Stay in the room when the problem resists. That resistance is where the understanding forms, and there is no shortcut to it, because the shortcut is what produces documents that look right and decisions that don’t hold.
If you liked this then you’ll love my other posts.