Will AI Make Specialists Unnecessary?

Probably not, if the last four technology waves are any guide. Each one made production cheap and left judgment exactly where it was, and the people who did well were the ones who already knew what good looked like.

I have been in the room for four of these now. Desktop publishing arriving at a newspaper. Non-linear editing arriving at a television station. The web turning every business into a software business, whether it wanted to be or not. And now this.

Each one showed up with the same three claims attached: it will make the work faster, it will make the specialists unnecessary, and if you do not move now you will be left behind. Two of those claims were usually wrong. The third was usually right, but never for the reason anyone gave at the time.

When desktop publishing landed in newsrooms, the promise was that anyone could lay out a page. Anyone could. The pages got worse for about three years, and then they got much better than before. Not because the tool improved, but because the people who understood typography learned the tool, and the people who only knew the tool did not learn typography.

When non-linear editing arrived, the promise was that editing would stop being a craft. It did not. What changed is that the cost of trying a version dropped to nearly nothing, so the good editors tried forty versions instead of three. The gap between a good editor and an average one got wider, not narrower.

When the web made everyone a software business, the promise was that off-the-shelf tools meant you would never need custom systems. Then I spent 2010 to 2013 building a custom ERP, inventory allocation and data-sharing across 57 retail stores for a jewelry chain, because the off-the-shelf answer did not know anything about how that business actually worked. And from 2013 I spent eight years building an insurance platform from nothing for a small brokerage, for the same reason.

That platform is the clearest evidence I have for any of this. It was not valuable because it used impressive technology. It was valuable because it encoded, correctly, how one particular business ran, to the point that it became part of why that business was worth acquiring. (The longer version of that history is on my About page.)

I use these tools daily. Not as a research interest, but as the way a meaningful share of my work now gets done, in a job where the results have to survive contact with real systems and real money. (Here is what that looks like on an ordinary day.)

What I notice is that the pattern holds, harder than before. The cost of producing a plausible answer has gone to approximately zero. The cost of knowing whether the answer is right has not moved, and in some places has gone up, because the plausible wrong answer now arrives formatted, confident and fast.

That is not a reason to avoid it. It is a reason to be clear about which half of your job it is doing. If your work is mostly production, this will change your work enormously and you should get very good at it, immediately. If your work is mostly judgment, it will change your day considerably and your value not at all, as long as you keep exercising the judgment rather than delegating it and hoping.

I do not think this wave is the same as the others. It is faster, it is broader, and it is the first one where the tool produces something that looks like judgment. That last part is genuinely new and I am not going to pretend otherwise.

But I have watched four of these arrive, and every time the people who did well were the ones who already knew what good looked like. That has not stopped being the thing worth having.