The order book is full. The bench isn't.

Two documents landed on my desk the same week. One put aerospace parts manufacturing on a path to $235 billion. The other listed the sector's hardest roles to fill – quality engineers, inspection and NDT-adjacent specialists, supplier-quality people – exactly the roles that decide whether $235 billion of parts are worth shipping. Every plant in the industry is advertising for the same person: five years of experience, sector stamps, available immediately, at a salary a mid-market supplier can actually pay. That person is already employed. Usually by a competitor. Not leaving. The arithmetic of the next decade does not close, because the cohorts that would be today's five-year veterans were never formed – five years ago everyone was hiring from the same empty pipeline instead of building one.

You don't hire competence – you manufacture it

I learned this in empty rooms. In 2019 I stood up a QA/QC department for a greenfield plant – SNOP, 900-plus employees, walls freshly painted, nobody in the building yet qualified to release a part. The labour market offered me enthusiasm. Enthusiasm is lovely; it cannot disposition a suspect lot.

So we stopped treating hiring as the acquisition of competence and started treating it as intake of raw material. Every inspector and quality engineer passed written qualification gates – the same thinking you apply in PPAP, aimed at a person instead of a part. Acceptance criteria were explicit: a seeded-defect test of forty parts, pass at 95% detection with zero critical misses, retested at fixed intervals. Nobody advanced a gate on tenure or charm.

Interviews, I can report with some embarrassment, predicted nothing. The candidate who talked the best game in a suit could not hold an argument about a caliper; a quiet woman hired off the line became the best auditor I had. Selection on evidence, qualification on measurement. That is the whole trick.

Standard work for judgment

Judgment looks like the one thing you cannot standardise. It isn't. It is badly documented, and we used the quality machinery itself as the teaching machinery.

QRQC is the fastest classroom on a shop floor: a defect found, fifteen minutes at the part, five whys with the person who found it. A novice who watches a real cause emerge from a real part learns more in that quarter-hour than in a week of slides. A3 works the same way – the left side forces a beginner to separate what they observed from what they assumed, which is half of this profession.

Boundary samples cover what words cannot. Nobody has written a usable work instruction for "acceptable porosity", but you can put the limit parts in someone's hands. Layered audits doubled as on-the-job qualification: each layer audited what it was competent to see and learned what the layer below looks at. Across a 2,000-plus, multi-site workforce we ran one A3 format everywhere, so a problem-solver trained in one plant was legible in every other.

And we error-proofed the novice the way you error-proof a station. An unqualified inspector could hold product but never release it. Drawing revisions were locked at the printer rather than trusting anyone to "be careful". Two weeks of supervised disposition preceded any independent sign-off.

You cannot recruit your way out of a competence deficit – you can only produce your way out.

The audit trail for people

Here is the test I apply to any plant's training system: show me an escape, and I will show you the binder that proves everyone was trained. A signature records attendance. It does not measure ability.

We audited training records the way auditors should treat device records, and we verified effectiveness as a number – escape rates before and after each qualification gate, drift checks at requalification. Visual inspection is a measurement system. The inspector is the instrument, and instruments get MSA, calibration and maintenance. Run an attribute agreement study on your inspection bench and you will learn which of your "experienced" people are actually repeatable; most plants have never run one.

The de-escalation years taught me the same lesson from the customer's side. The quarter we ran with zero critical customer escalations was not the quarter with the most inspectors. It was the quarter in which people were trained to escalate and disposition on evidence. Trained judgment beats headcount.

Run the people system like a production system

Competence has yield, scrap and cycle time whether or not you measure them. Yield: the share of hires reaching fully qualified status within the planned time. Scrap: the failed hire – recruitment fees, six months of salary, training hours, then a restart – comfortably €50,000 for one quality engineer, before counting the escapes they signed off. Cycle time: weeks from day one to independent disposition authority. Those three belong on the same board as OTD and PPM, because they sit upstream of both.

The arriving generation of AI agents does not dissolve this problem; it sharpens it. Automate the routine checking and you also automate away the accidental apprenticeship – the junior years spent doing simple checks next to someone who knew why they mattered. The judgment layer that must challenge the machine when it drifts gets scarcer, not cheaper. It is not on the market at any salary the mid-market can pay. It has to be produced.

The order book is full everywhere. The bench is only full where someone decided to build one. The scarce skill of the next decade appears on no CV, because a CV records where a person has been, not what a plant produced in them. The winners give competence a routing, a takt and a scrap rate – and hold it to the same acceptance criteria as the parts it releases.