You're classifying the messy middle
This is one result from the AI Job Drift Diagnostic. It means your misses are mostly on tasks that do not have a clean answer. The artifact part of the task has moved to AI. The context part has not. You are landing in the honest territory: work that splits in the middle and requires a different kind of ownership than either pure execution or pure judgment.
Why some tasks resist clean classification
Most tasks have two layers. The artifact layer is the thing you produce: the report, the draft, the summary, the response, the plan. The context layer is everything that makes the artifact mean something: the stakeholder reading it, the history behind the situation, the stakes if the output is slightly wrong, the unspoken brief underneath the visible one.
For most of professional history, you could not separate these layers. Producing the artifact required owning the context, because the artifact only existed through sustained human effort. AI changed that. The artifact layer has become detachable. AI can produce a credible first pass at a brief, a report, or a summary without holding any of the context that makes those outputs matter.
This creates a genuinely new category of work that did not exist five years ago: tasks where AI owns the production and you own the framing and the judgment around what gets produced. That is what boundary work is. Neither pure execution nor pure judgment. A new kind of collaboration.
What boundary work looks like in practice
A customer complaint. AI reads the account history, classifies the sentiment, and drafts a response that is technically accurate. Your job is to notice that this particular customer is not angry about the immediate issue — they are angry about the pattern, and the technically accurate response will make it worse. AI produced the artifact. You held the context that determined whether to use it.
A strategy document. AI synthesizes the research, formats the options, and drafts a recommendation that is logically coherent. Your job is to recognize that the recommended option assumes a relationship with a partner that is about to change in ways the data does not capture yet. The synthesis was real work. The judgment about what to do with it was still yours.
A code change. AI writes working code for a feature. Your job is to recognize that the implementation, while correct, will create a maintenance burden in six months that is not obvious from the ticket. The execution moved to AI. The architectural judgment did not.
In each case, the task is genuinely shared. Assigning it entirely to AI loses the context layer. Keeping it entirely to yourself wastes the production efficiency AI provides. Neither extreme is right.
How to work the messy middle
Split the task explicitly, not implicitly. When you approach a boundary task, decide in advance what you are delegating to AI and what you are keeping. "AI drafts, I judge" is a different workflow from "AI drafts and I clean up." In the first, you are evaluating the output against your context. In the second, you are editing the output on its own terms. Only the first captures the value you actually provide.
Keep the brief, not the draft. Your irreplaceable contribution in boundary work is usually upstream: the framing of what the artifact should accomplish. If you hand AI both the brief and the production, you have handed the whole task. If you write the brief yourself and hand only the production, you keep the part that is actually hard to replace.
Trust the output more, not less. One failure mode in the messy middle is editing AI outputs down to something that looks like what you would have written anyway. That recaptures execution work without adding judgment. The question is not "does this sound like me?" The question is "does this serve the context correctly?" If it does, use it. The goal is not your voice. It is the right outcome.
Why this is the most honest result
The cleanly aligned result means the tasks are easy to sort: execution here, judgment there. The boundary result means you are working in territory where the sorting is genuinely hard and most people get it wrong.
People who over-protect boundary work treat the artifact as the value and hold the whole task. People who under-protect it hand the brief along with the production and lose the context layer. The right position is narrow: own the framing, evaluate the output, and let AI carry the artifact. That is more demanding than either extreme, because it requires knowing exactly what layer you are adding on each pass.