The most common claim about AI and professional work is also the least useful one: that it will replace expertise. Watch what actually happens in a working week and a different picture appears. AI is very good at the middle of the work and unreliable at both ends.
Where the work actually breaks
Professional work has three phases:
- Framing. Deciding what the real problem is, what the constraints are, what can be ignored.
- Production. Drafting, generating options, moving material into a usable shape.
- Judgement. Deciding whether the output is correct, safe, and worth someone’s time.
AI compresses the middle phase dramatically. It does not touch the first phase, and it actively increases the load on the third. When production becomes cheap, the volume of material needing judgement goes up, not down.
Why the first draft was never the hard part
In architecture and urban planning, the expensive failure was rarely the drawing. It was the brief. A beautifully produced scheme that answers the wrong question costs more than an ugly one that answers the right question, because the error is discovered later and by someone else.
The same is true of an AI-assisted article or an AI-assisted website. The model will produce something coherent within seconds. Whether it is right depends entirely on whether the person asking knows what right looks like in that context.
What this means in practice
Three habits have made the difference in my own work:
- Write the brief before opening the tool. If I cannot state the constraint in one paragraph, I am not ready to delegate any part of it.
- Keep a rejection log. When I discard AI output, I write down why. That log is now the most valuable document in my workflow, because it encodes judgement that no prompt captures.
- Stay the author of record. If I cannot defend a sentence in a meeting, it does not ship under my name.
The part nobody wants to hear
This shift is bad news for anyone whose value was located in production speed alone, and good news for anyone who has spent years building judgement. Experience does not become irrelevant. It becomes the only thing that is scarce.
The risk is not that AI replaces experts. The risk is that people stop doing the reps that make someone an expert, because the first draft no longer forces them to think.
