LANDMARK · 7 MIN · JUDGE
When not to automate
After this landmark, you can recognise the task shapes where automation is the wrong answer even when the tool performs well, and say why in terms someone can act on.
Most advice about AI adoption is about where it works. The more useful skill is recognising where it shouldn’t go even when it works. Four shapes are worth memorising. First, tasks where being wrong is not recoverable — a payment sent, a message published, a record deleted. Second, tasks where the process is the point: a condolence note, a performance conversation, a decision someone is owed an explanation for. Automating those doesn’t just risk a bad output, it removes the thing that gave the output meaning. Third, tasks where you cannot check the answer — if you have no way to verify the output and no source to compare it against, a fluent answer is worse than no answer, because it displaces the uncertainty you should have felt. Fourth, tasks where the errors land on someone other than the person deciding: applicants, patients, claimants, students. A 95% accurate system is a good tool in the first case and an unacceptable one in the fourth, with no change in the technology.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
Don’t let an assistant write the message that matters to a person who knows you. Use it to think, and then write the thing yourself. And if you can’t check an answer — a medical dose, a legal deadline, a tax rule — treat a confident response as a prompt to consult a real source, not as the source.
MAKE A DECISION
A hiring team wants an AI system to rank applicants and auto-reject the bottom half, with recruiters reviewing only the shortlist. What is the sound judgment?
CARRY THISWrite down one task in your work you have decided not to automate, and the reason in one sentence: unrecoverable, the process is the point, unverifiable, or the errors land on someone else.