Your Hardware Talent Was Always the Bottleneck

The Individual Contributor Age was a software story: engineering capacity stopped being the bottleneck; the person who knew what was worth building became the scarce asset. Hardware organisations still behave as if the scarce asset is the few people allowed to touch NX, the PLC, the qualification packet. That was true when implementation was a year. It is less true when a closed loop can cut validation from four days to two, when a schematic can write its own regressions, when a sentence can become a machine that runs.^[1]

Your hardware talent was always the bottleneck. We just blamed the tools. The tools caught up. The org chart is still rationing the humans who hold the exception, the material, the history of the lot. That is the expensive mistake of the next three years.

The IC on the bench already had the roadmap

Ask a process engineer, a controls tech, a supplier-quality lead, a designer who still opens the assembly what they would build if they could. You will get a coherent backlog in thirty minutes. The unofficial MES notebook. The pinout that is true. The fixture family. The twin-diff that should have been CI. None of this was a mystery. It sat behind whoever was funded to touch the official stack this quarter.

That gate produced the industrial org we know: a small priesthood, a long ECO queue, a central digital function whose job was to triage the gap. Same geometry as the monolithic product org. AI is the leveller here too, not magic, compressed implementation. When a serious operator plus a skill folder plus a bench can close a loop in days, waiting for the priesthood is no longer governance. It is waste with a safety poster.

What gets scarce

The same three things as in software, with mass:

• Clarity, what "done" means on this lot, this cell, this customer.
• Taste, cathedral vs tangent, DfM that will cut, an HMI a technician will trust.
• Memory, why the exception exists, which field in MES is a lie, which allowable was a one-time deviation. Models do not have that unless you put it in a skill.

Implementation is getting cheaper. Those three are not. A senior process engineer with a library outruns a transformation PMO. A junior with ChatGPT and no floor is how you get a ban. Empowerment without a lab next to the line is how juniors get hurt.

The filter moved up the stack

When everyone has Claude Code, what filters talent? The hardware version: when everyone can spawn geometry, who do you promote? Not the person who produces the most variants. The person who:

• Falsifies cheaply on the bench, a failed eval is a gift, not an insult.
• Authors a skill from that failure, description that triggers, body under 500 lines, script that exits 1.
• Knows which objects are allowed to die.
• Owns the loop after the FDE leaves.

The IC who versions skills is the new staff engineer. The IC who collects prompts in a document is the old one. Hire and promote for the first. You will be tempted to hire "AI people" who have never stood next to a cell. That is how you get twin theatre.

Give them the stack now

Talent this valuable does not wait for the operating model. Managed runtime, role bundle, classified slice, bench time, a stop they will hit. Career path: operator-in-residence → technical operator back on the line → FDE or platform. The lab is not a parking lot for people engineering did not want. It is how you grow the IC you actually needed.

Compensation should not look like a hackathon T-shirt. If the unofficial notebook already runs the region, paying them like a technician while you hire consultants to "digitise" it is how you lose them to a Hadrian.

Failure modes

Priesthood as identity. Only the named engineers may prompt. Everyone else pastes anyway.

AI people instead of domain people. Pretty twins. Yield unchanged.

Champion network. Spare time, town-hall volunteers, no picture.

Tools next year. Drawings in consumer chat this year.

No path back to the line. You trained them and parked them in a lab campus. They leave.

Hardware talent was always the bottleneck. Implementation was the alibi. The alibi is dying. Hunt the ICs who already cheated toward the work, give them folders and a floor, filter for the mindset that can turn a failed lot into a skill. The companies that look like they "adopted physical AI fast" will have done something ruder: they stopped rationing their own people.

Related: The Individual Contributor Age, The Battle of Mindsets, and The Lab Next to the Line.