The Business Times ran a piece on 2 November 2026 about manufacturing’s “black art”, the know-how that sits in the heads of veteran machinists and engineers who are now retiring. Its argument is that AI may be the best chance to keep that knowledge before it leaves with them. The summary line puts it plainly: manufacturing may not be the most glamorous sector, but it could be where AI proves its real worth.
The need is real. The assumption under it is not.
I agree with the need. Most businesses have one or two people who carry the operation in their heads. When they go, the business gets slower and makes mistakes it never used to make, and nobody can say exactly why.
The assumption I don’t accept is that AI absorbs this by exposure. Put a model near the machine, feed it the logs, and the skill transfers. That is the hype version, and it rests on nothing.
Here is what actually changed. A machine can now take a description in ordinary words and turn it into instructions it runs the same way every time. That is real, and it is new. Here is what did not change. The knowledge is still inside the person. A model trained on sensor data gets the numbers. It does not get the reason the machinist eased off at that moment, because that reason was never recorded anywhere.
Watching gives you data. It doesn’t give you the rule.
A machinist hears a cut going wrong before the gauge shows it. Ask them how and they say “it sounds wrong”. That answer is useless to a model and useless to a new hire. The useful version takes an hour of patient questions. What did you hear? What did you check next? What did you do? When would you have stopped the machine instead of adjusting it?
That is a briefing. It is the same work whether the thing receiving it is an apprentice, a procedure document or an AI. I teach a principle that applies here: context is the product. The quality of what goes in decides the quality of what comes out. The technology is the easy part. Getting a person who has done something by feel for thirty years to say it in words is the hard part, and it needs a second person who knows how to ask.
Manufacturing is the clearest case because the expert is literally retiring. The same gap exists in every service business. The clinic receptionist who knows which patient to squeeze in today. The shop manager who knows which complaint is about to become a chargeback. The one person on a marketing team who knows why a client always rejects the first draft. None of it is in the handbook. All of it leaves when they do.
The briefing is the product. I see this every day.
With AeroChat, a merchant describes the workflow they want automated in their own words, and the system translates it into instructions and runs it. The description is where the work is. A vague description produces a vague workflow. A detailed one, with the exceptions the senior person would have caught, runs the way that person would have run it. The model does not close that gap on its own. The merchant closes it by saying what they actually do.
My own working rule is brief, produce, check, then fix the brief. When the output is wrong, the first place I look is what I told it, not the tool. Almost every time, the brief was missing a condition that was obvious to me and invisible to the machine.
What to do before the expert leaves
Start the briefing now, before you buy anything.
- Name the person. Pick the one whose retirement would hurt most and give them time off the floor to do this.
- Record decisions, not steps. Steps are already in the manual. Ask what they look at, what tells them something is wrong, what they do next, and when they would stop and call someone.
- Separate rules from judgement. Tolerances, stop conditions and safety limits are rules, and they go into the system as hard limits. Judgement calls stay as described situations with the usual answer. Rules where wrong is costly, AI where judgement is needed.
- Test the written version. Give it to a new hire or to the AI and watch where it fails. Every failure is a missing line in the brief. Add it and test again.
- Keep the expert in the loop while they are still here. They are the only person who can tell you the written version is wrong.
The model you pick will change within a year. The briefing you get out of your retiring machinist will still be true.
Source: Business Times: Manufacturing’s black art is retiring. AI may be the best chance to keep it