Someone asks for the top five risks in the supplier agreement. They get five. All five are real, well described, correctly drawn from the document. What mattered was a termination provision that is not a risk in the ordinary sense: it is a commercial constraint that changes what the business can commit to next year. Nobody asked, so nobody got it.
The answer that came back was excellent. That is the whole problem. A model will answer the question it is given, and a fluent answer to the wrong question is indistinguishable, on the page, from a fluent answer to the right one.
So where did that go wrong? Not in the extraction. The extraction was perfect. It went wrong one step earlier, in deciding which clauses mattered enough to extract, and that step looks like procedure and is not.
Professional work divides into two modes, and almost every piece of real work contains both.
Procedure is rule-governed work. There is a right answer, there is an external standard, and there is a way to establish whether the step was done correctly. Extracting the dates, fees and notice periods from the contract. Checking the clause set against a standard checklist. Building the cost comparison from the invoices. You do not need to be senior to check this work; you need the rule and the source.
Craft is judgment-governed work. What makes it good is contextual, much of what makes it good was never written down, and two competent people can legitimately disagree about it. Deciding what this renewal actually turns on. Judging whether the switching risk is real or boilerplate. Knowing that the two integrations nobody documented are the reason leaving this supplier is harder than it looks. There is no external standard for any of that.
Each mode fails when it is handled with the other's materials, and the failures are expensive in opposite directions. Standardise craft and you get compliance without judgment: the template filled in, the boxes ticked, the thinking gone. Leave procedure unstandardised and you get inconsistency at scale: forty renewal notes a year, each checking different things, none comparable. An organisation that cannot tell which mode it is looking at will reliably commit both at once.
Now the trap in the title. Take a piece of work apart and most steps fall clearly on one side or the other. The dangerous ones are the steps wearing the other mode's clothes, and they run in one direction: a judgment step that presents itself as a rule. Deciding which clauses matter. Deciding which five risks. Deciding what counts as an exception worth escalating. Each of those has the shape of a rule somebody could write down, which is why somebody eventually does, and the writing down is where the judgment leaves.
Three cautions, because the distinction turns useless without them. Mode is not seniority; senior people do plenty of procedure and juniors exercise real craft, and mapping mode onto grade is the fastest way to waste the idea. The boundary moves, and mostly in one direction: work that was craft becomes procedure when the standard finally gets articulated, which is what a maturing profession does. And craft is not a shelter. "This is craft, so it cannot be assessed" is exactly what a professional who does not want to be assessed will say, and I am not handing them the words. Craft is assessable: by demonstration, by calibration against peers, by whether a practitioner can explain why a piece of work is good rather than that it looked fine. Just not by examination.
The decomposition is the work. Point the machine freely at the procedural steps, where the rule and the source make checking cheap. Treat the craft steps as yours, with the machine arguing against you rather than answering for you. And when a step seems to be both modes at once, split it rather than average it. Averaging gives you a step that gets checked like procedure and judged like craft, which means it gets neither.
This is a working distinction rather than a research finding; it came out of watching where AI-assisted work goes wrong, not out of a study. That the two modes fail differently, and that treating one as the other is where the expensive mistakes live, I would defend anywhere. The clean split is another matter. On a page it looks tidy. In real work there are steps I still cannot place.
Something to try this week. Take one piece of work you do repeatedly and write out its steps, ten or twelve of them, in the order you actually do them. Mark each one P or C. The exercise takes twenty minutes and most of the marks are obvious. Find the step you argued with yourself about. That is the one to look at before you automate anything.
The uncomfortable version. Give the same unmarked list to a colleague who does the same work, and ask them to mark it independently. Then compare. Where the two of you disagree is where your team is about to standardise something it should not, and until now neither of you knew.