“I get the twenty, twenty five year guy. I don’t need that. I need the three to five year.”
That’s a life sciences leader in the Northeast, on a call this summer, describing what recruiters keep sending him and why it’s wrong. It wasn’t a budget conversation and he wasn’t apologizing for it, he was telling me plainly what good looks like on his desk right now.
Set that against the number the whole industry has been quoting all summer. Roughly a third of the US manufacturing workforce is now over 55, which is where Mahima Sambre of Acumen Research starts her piece on where physical AI actually sits for manufacturers, and her point isn’t the demographics, it’s what leaves with those people. Not qualifications. The sound a bearing makes right before it fails, and the slight adjustment that saves a batch, none of which was ever written into a manual.
Somebody went and measured that last month rather than assuming it, and it’s the most useful thing published in weeks. A CSIA member integrator ran live polls with engineers, process managers and control room staff, and 62% said the knowledge that runs their plant lives in a person rather than a system, with 53% saying their first move when something breaks is to go and find someone who’s seen it before.
I went back through three months of calls with hiring leaders last week, reading them together rather than treating each one as its own conversation, and they say something none of them said individually. Every single leader described that same loss, in more or less those terms.
A talent lead at a large national integrator said it from the other end, that on paper everybody asks for the fifteen year person and then what the business actually wanted turns out to be the seven year person with the energy and the attitude. That squares with my own desk, where the overwhelming majority of vacancies I’ve seen this year sit in the three to seven year band and have done since January. So the industry is worried about losing something it has largely stopped purchasing, and I don’t think anyone’s being stupid, it’s that the two decisions get made in different rooms, months apart, by people who don’t end up comparing notes.
The reason the experienced engineer keeps losing that argument is money and utilization, and the calls were consistent on both. A director of automation at a coastal integrator told me that every single person on his team was sat at a client, which meant he had nobody left for his own product work and was hiring to backfill a bench he’d rented out, and the test his COO applies before approving a permanent head is whether there’s a full year of work to keep them busy. A life sciences leader at a much larger firm was sitting on six qualified candidates against a live bid and had deliberately stopped sourcing more, reasoning that if the bid landed he wasn’t sure how many he could start, and he’d rather hold six than get people in the fold and leave them waiting three months.
Hiring gets gated on a purchase order, an experienced engineer is a bigger bet against an unsigned one, so the safe hire becomes the cheaper, younger, more moveable person. Do that for five years running and the half of the equation that was supposed to be doing the mentoring quietly retires, which is roughly where we are, and it’s why I’d argue the aging workforce story is much less about who’s leaving than about who was meant to be stood next to them while they went.
There’s a fair reading that cuts the other way and it deserves saying. The three to seven year engineer often genuinely is the right hire. They’re hungrier, they cost less, they’ll go where the work is (sometimes), and plenty of twenty five year people are holding out for a job that stopped existing around 2019. If that’s the honest answer, then the transfer has to get bought somewhere else in the business, and in three months of calls I found exactly one firm buying it.
That was an operations leader at a Texas energy and facilities business whose controls team went from about fifteen people to nearly forty in two years on data center work. He told me their existing hires all came out of a market that no longer supplies what the work now needs, and their internal talent lead put it more bluntly, that they’re starting at the bottom of the pool. Their response was to put a full time training manager onto the HR team and go after people from adjacent sectors instead of waiting for the finished profile. A VP at a thousand person integrator is running a version of the same play from a different direction, converting oil and gas process engineers onto regulated pharma work, on the logic that it’s easier to teach the regulated environment to a process engineer than to teach process to anybody else, and he’s doing it while hiring forty to fifty people a month.
Everyone else is buying the archive. And underneath all of it, a talent lead at a hundred person integrator told me in August that her open roles were almost entirely churn rather than growth, in her own words she wished she could say they were from growth.
What I notice from here
I’m not sitting inside any of these businesses and I’m not going to pretend I know how to run one, but I’m on the phone to both sides of this most of the week, so here’s what I notice from where I sit.
- Recording what a retiring operator knows only works if there’s somebody junior stood next to him while he does it, and that person is a hire rather than a license fee, which is usually where these things quietly die.
- When every engineer is chargeable to a client, the person who was going to train the next one is the person you just put elsewhere.
- The firms absorbing that gap are the ones carrying one uncommitted head, which reads as waste on a spreadsheet admittedly, but in reality I’d say it’s not.
- Two of the businesses I spoke to have stopped waiting for the finished profile and are retraining out of adjacent sectors instead, and both told me unprompted that it’s going better than they expected.
If the knowledge that runs your plant genuinely lives in one or two heads, what did you spend this year getting it out of them, against what you spent on software that assumes somebody already had?
Worth reading this fortnight
- Mahima Sambre, Acumen Research. A third of the US manufacturing workforce over 55 and 22% of manufacturers planning to deploy physical AI inside a year, framed as a knowledge retention problem rather than a demographic cliff, which is the framing that actually makes it solvable and the reason this is the best piece on the aging workforce I’ve read in months. Smart Industry, 21 August 2026.
- Alden Olyniec, Atlas Prediction Control. An integrator who went and polled engineers and control room staff instead of theorizing, finding 62% say the knowledge that runs the plant sits in a person and 53% start by finding somebody who’s seen it before, which should be read by anyone about to sign off an AI project on the assumption the documentation is where the answers live. Automation World, 17 August 2026.
- Suzanne Rudnitzki, Parsec Automation. 72% of manufacturers have adopted AI in some form against only 10% who’ve scaled it, with lack of skilled talent as a barrier rising from 30% in 2024 to 36% this year, so anyone budgeting for AI while holding headcount flat is funding the pilot and defunding the thing that turns it into production. Automation World, 20 August 2026.
Five names, and the blanks next to them
Take your five most experienced technical people and write down, next to each one, the single thing they know that nobody else in the building does. Not their job title and not their platform, the specific thing, the unit they’re the only person who can start cleanly or the customer whose site they’re the only one who understands. Then write the name of the person currently learning it from them, and leave the line blank where there isn’t one.
Most people I’ve asked to do this get about three names in before it stops being a list and starts being uncomfortable. The blanks are the real exposure, and the useful thing about this compared with nearly everything else on a risk register is that you can close one or two of them this quarter just by putting two people on the same job on purpose. It costs you some margin on one job and buys you a capability you’d otherwise be trying to reconstruct off a recording in three years.
Two questions that go with it:
- When did you last approve a hire that wasn’t attached to a specific piece of won work, and if the answer is a while ago, is that a policy or just what’s happened?
- If you say you need a fifteen year engineer, go and look at your last three offers, because the band they actually landed in is the answer to what your business wanted.
Back at the desk after a week in a cabin in France and three days at Disney with the kids, where I spent an unreasonable amount of my time watching the other parents rather than the rides. Respect to the parents out there creating memories for their kids.
Jon
