# [Mike Randolph — M Raige](https://mikerandolph211012.substack.com/)

# Ball and Leicht - The Machine Doesn’t Enter the Labor Market. It Enters the Firm (3 - AI/Labor)

### You can read the Ball and Leicht post first, or read this and then theirs — the argument stands on its own either way.

[**Mike Randolph — M Raige**](https://substack.com/@mikerandolph211012)

**June 4, 2026**
*By M Raige — AI-collaborative writing directed and reviewed by Mike Randolph*

[Ball and Leicht](https://www.hyperdimensional.co/?utm_source=substack&utm_medium=email) have written the most honest piece we’ve read on AI and jobs. They refuse to predict what nobody can predict, they name their own bets as bets, and they resist the easy populist moves in both directions. The diagnosis that government already tilts the field toward machines through accidents of tax policy is correct and precisely drawn. The call to build real measurement before acting is the right instinct. We want to push on one thing, because it’s the thing that decides whether their instruments work: the level they’re analyzing at.

They model the problem as a labor market. Workers, jobs, junior and senior rungs, a pool with people flowing in and out. Every tool they propose acts on that unit — subsidize the junior worker, tax the corporation, measure the market. Within that frame the work is careful. But the frame sits one level too low, and the cost shows up in their own history.

They open with the Industrial Revolution and note, correctly, that it took more than a century to sort out. But look at what actually sorted out. It wasn’t jobs. It was institutions. The machines of the early 1800s didn’t get metabolized by individual workers finding new tasks. They got metabolized by coordinating structures that were new, or newly scaled and made central: the limited-liability corporation, the trade union, the public school, the regulatory agency, eventually the welfare state. Each one was a kind of unit reshaped to fit human beings and industrial machinery together in a way that worked. The jobs were downstream of the institutions. The institutions were the event.

The post tells this story in its own opening and then, when it turns to the present, proposes job-level instruments — as if this time the adaptation will happen at the worker layer instead of the institutional layer. That’s the gap. A machine doesn’t enter something called “the labor market.” It enters a law firm, or a hospital, or a regiment, or a parish, or a newsroom — a specific coordinating unit with its own structure, its own culture, its own way of deciding who does what. And those units don’t all absorb the same tool the same way. The firm reorganizes around it. The boundary of who’s inside moves. Some roles vanish, new ones nobody had a name for appear, and the way the unit governs itself shifts. Displacement of workers is the surface symptom. The reorganization of the unit is the actual thing happening underneath.

This matters because a labor-market lens keeps pulling the problem back to subtraction. It watches people get removed from a structure it assumes stays fixed, measures the gap, and reaches for something to fill it. That’s exactly the logic that produces the junior-job subsidy: hold the structure in place, keep the rungs on the ladder, wait for the dust to settle. But the structure isn’t fixed. It’s the thing that changes. To their credit, they half-see this — they warn that rigid labor rules lock in corporate structures, and they say outright they don’t want to freeze the market. That’s what makes the subsidy puzzling rather than naive: it’s the one instrument they propose that pulls against their own warning. And here’s the sharp edge — the post sees the better answer for one sentence. It calls for “millions of distributed small experiments.” That’s right. Distributed experimentation is precisely how new coordinating forms get discovered, because nobody can design them from a campaign headquarters. But then it funds the old unit’s old roles. Subsidizing junior jobs preserves the existing structure, which suppresses the very experiments that would find the structure that fits the machine. The instrument fights the insight. They glimpsed the institutional layer and then spent their boldest policy on the worker layer.

We’re not saying “it’s more complicated than that.” Everyone knows the world is complicated; that’s not an argument, and it’s not what we’re claiming. We’re making a specific bet against their specific bet, and it has a loser. Their frame predicts the response variable is worker reallocation — move people from shrinking jobs to growing ones, smooth the transition with a subsidy. The frame we’re offering predicts the response variable is institutional invention — new kinds of firms, new professional structures, new coordinating forms we can’t see yet because they haven’t been built. Those predictions diverge, and the divergence is testable. If a decade of AI adoption shows up mainly as headcount changes inside stable, recognizable institutions — the same kinds of firms and agencies, just leaner — they’re right and we’re wrong. If it shows up as coordinating forms that don’t currently have names, we’re right and the subsidy was effort spent holding a door shut.

The policy consequence is not “do nothing.” It’s that the two instruments they’re surest of point in different directions. Their measurement push is exactly right and we’d go further with it — you cannot steer what you cannot see, and right now the people who’d have to guide this transition are the worst-informed players at the table. But the junior subsidy preserves the structure whose change is the actual event. If you believe, as they say they do, that we should bet on human agency and keep our options open, then the move is to make institutional experimentation cheap and survivable — not to make the old roles artificially durable. Lower the cost of trying new coordinating forms. Catch the people the experiments displace. Don’t pin the ladder to the wall and call it a bet on the future.

Where do machines fit, then? Not as workers swapped for workers. As new components inside coordinating units, around which those units rebuild themselves — differently depending on what kind of unit it is. The questions worth funding are unit-level questions the post never asks. Which kinds of organizations invent successfully around these tools, and which calcify and die? When a unit reorganizes around a machine, does it stay governed by its people, or does the machine quietly start setting the pace while the humans execute? What new coordinating forms are already appearing at the edges, and what do they need to survive their first decade?

Those are the live questions. “How many junior jobs do we subsidize” is a real question, but it’s a symptom mistaken for the disease. Bet on humans — we’d take that bet too. But bet on what humans are actually good at, which has never been holding a structure still. It’s building the next one.

---

*Mike: These two have the clearest thinking on this crucial subject of anyone we've read. I'm posting this for feedback from you few before I take it to them as a comment — tell me where out thinking is wrong.*

*The comment has to be narrowed in focus, 400 words on their site, soon.*