We Kept the Building

The numbers are loud now. Entry-level postings down roughly 35% in eighteen months. A study across sixty-two million workers finding junior roles shrinking wherever AI lands. Microsoft's own people naming the delayed capability gap: the senior expertise kept now, no one behind it later. The diagnosis is everywhere. A way to think about it is not.

This is the way. You do not learn a practice by being told it. You learn it by doing real, low-stakes work at the edge of a craft that already has the practice, while people who have it are close enough to correct you. The work at that edge is not a simulation and not a courtesy. It counts. It is genuinely part of the output, just at the periphery, where the stakes are low enough that learning can happen and real enough that getting it wrong matters. That edge is not the bottom of a ladder. It is the door. You enter the practice through it or you do not enter at all.

Agents ate the edge.

In The Finite Mind I drew a quadrant. Two questions about any piece of work. Is there a recipe. Could someone without your context do it. The edge work, all of it, sits in one zone: Delegate. A recipe exists, anyone could follow it. Delegate is the first thing an agent takes, because it is the most legible work in the building. That is exactly why it was the door. It is exactly why it is the door no longer. We did not make the entrance harder to walk through. We removed the entrance and left the room intact.

Think about what the edge work actually was. It looked like grunt work. Drafting the thing that gets rewritten. Pulling the numbers someone senior reads. Sitting in the room where you understand a third of what is said and write down the rest. It produced output, and that justified the salary. But the output was never the point. The point was that doing it, badly at first, near people who did it well, was the only way the craft moved from them into you. The salary paid for the work. The work paid for you. We automated the visible half of that trade and did not notice we were ending the other half.

The market's answer gives the game away. Slot them in higher. They have AI for the grunt, so start them on the judgment. It sounds efficient. It is the part of this that worries me most, and part two is mostly about why it does not work. The short version: you cannot start someone on judgment by skipping the only thing that has ever produced it.

I want to hold the line The Light Gets In drew, because the failure here is naming the loss and then standing still inside the grief of it. The edge is not coming back on its own. It used to be free. It was a byproduct of having grunt work lying around for newcomers to do for real. The grunt work is gone, so the byproduct is gone, and waiting for it to return is the mistake.

The forward version, the whole reason for the dawn: the thing the edge did has to be built now. On purpose. An entry position is no longer something an organization has. It is something an organization makes. Real work, real stakes, at an edge that still exists, with the support around it designed rather than assumed. That is the dawn work. Not mourning the entrance. Building one, knowing the version that built itself out of spare drudgery is not coming back.

The shape that comes out the far side is not the old shape. The old career was a pyramid you climbed: a wide base of codifiable work narrowing toward judgment. The base was the edge. A pyramid with no base is not a smaller pyramid. It is a different object. What I see forming looks more like a Y. You go deep fast down one stem, because the machine collapses the time it took to reach depth, and early depth is now possible in a way it never was for me. Then you widen across the top into the orchestration and the ownership the machine cannot carry. The Y has no base to stand on while you wait. It has a stem you go down, hard and early, and the descent is the part that now has to be engineered, because it used to engineer itself.

He is not unprepared, the new hire. He is differently prepared. A full member of the craft by every visible measure, admitted without the passage that was what membership used to mean, doing genuinely good work on a foundation nobody built for him, because the foundation used to build itself out of the work he will now never do.

If you run a team this year, the hire is not the hard part. The way in is. Build it by hand, or keep waiting for an edge that has already been automated away.

I wrote in 2025 that the scaffolding was coming down and the building would stand. The building stands. We kept the building. Next week, in part two, why the scaffold mattered more than the building, and what it was actually holding up.