You will hear this at every conference and in every LinkedIn thread for the next decade: AI cannot replace real mechanical engineering. It cannot feel a tolerance stack go wrong. It cannot walk the floor and notice a fixture is drifting. It cannot sit across from a supplier and read the pause before the lie. All of that is true. None of it matters. Because the job you actually do every week is not that. See Reason #40.
You already read what the job became. You route ECOs through approval chains. You fill out DFMEA templates one failure mode at a time. You build DV/PV matrices in spreadsheets and track them in portals. You write test reports that exist to prove something passed, not to explain why it works. You chase RoHS and REACH certificates from suppliers who do not answer emails. You update BOMs in ERP systems that fight you. You reformat PDFs because a customer portal rejects embedded fonts. You paste screenshots into PowerPoint decks that a manager will skim for one bullet before asking for a risk line. See Reason #33 and See Reason #9.
That is the job. Not the brochure version. The calendar version. The version you live Monday through Friday, and often Saturdays as well. And every single item on that list is text-in, text-out work. It is structured, repetitive, and traceable, which is exactly the profile that large language models and workflow automation were built to eat. The question was never whether AI could replace a mechanical engineer who sizes a pressure vessel from first principles. The question is whether AI can fill out the paperwork that surrounds the pressure vessel after someone else already sized it. The answer is yes. It already can.
The compliance layer accelerates this. You spend increasing fractions of your week not designing but proving, assembling cert packs, mapping test evidence to requirements, building traceability matrices, and writing justification memos so an auditor can check a box. See Reason #51. That work expanded until it became the job. And it expanded into the exact shape of a task that automation handles well: collect inputs, apply rules, generate output, route for signature. You did not need to be replaced. You needed to be transcribed. See Reason #65.
The people who say "AI can't do what I do" are thinking of the 20% that still feels like engineering. The thermal intuition. The fixture hack that saved a build. The moment you overrode the model because you remembered a field return from 2016. That part is real, and no model replicates it today. But that 20% does not justify the headcount. The headcount was justified by the other 80%, the administrative throughput that kept gates moving, reports filed, and portals green. See Reason #42 and See Reason #26. When that 80% gets cheaper to automate than to staff, the headcount shrinks. You keep the title. You lose the seat.
This is not speculation. It is the same pattern that played out when admin work moved offshore, except faster and without the time zone lag. See Reason #40 already told you the rule: if the work can be written down, the work can be moved. Now it does not even need to move. It just needs a prompt.
The field will not vanish. Someone will still walk the floor. Someone will still argue with a casting vendor about why the draft angle cannot drop another half degree. But there will be fewer someones, and the ones who remain will be expected to carry the 20% that matters on a fraction of the old headcount, while a dashboard handles the rest. That is not survival. It is compression. And in a market that already has two and a half candidates for every opening, compression does not create opportunity. It removes it.
You were told your judgment makes you irreplaceable. It does. The job just stopped being about judgment a long time ago.
No comments:
Post a Comment