I had a messy job: senior automotive regulatory engineer. Negotiation was central to the job. For example, the Pedestrian Safety Enhancement Act of 2010 required the National Highway Traffic Safety Administration to develop a regulation to require electric cars and hybrid cars running in electric mode to produce synthetic noise below 20 mph (the speed above which a vehicle's tire and aerodynamic noises become conspicuous and predominant to nearby pedestrians).
Besides the technology challenges (the technology didn't exist in 2010), this would be a regulation that would affect every car manufacturer in the world. Then there was the worldwide FOMO effect: other countries enacted their own laws to ensure they were seen as proactive and to have a seat at the negotiating table. Some countries, like Germany, had antithetical environmental laws mandating the REDUCTION of vehicle noise. NGOs like the National Federation of the Blind were also involved. Inventing compliance testing machinery and methods was maddening. In the process, I learned to navigate neighborhood sidewalks in Baltimore blindfolded, and wore 3-piece suits to post-Fukushima (no air conditioning!) multi-day meetings in Osaka, Japan.
The final regulation - FMVSS 141 - went into effect twelve years later.... but is still being argued.
I have a question. Do companies use stripped down versions of AI for their service / support chats? Almost every one I have interacted with seems dumb as a post. They don't venture past a "scrpt", have no power of judgement, and seem inadequate at best.
You write that "junior roles are disappearing". If AI is dismantling the bottom rung of the career ladder, the the reduction in jobs should appear in the unemployment rate for 20 to 24 years-olds. It does not. The current reading of 7.2% sits below the 2016 average of 8.4%, a year when nobody worried about AI eating entry-level jobs. The 2025 rise to 9.2% looked like the start of something, but it has fully retraced. The first five months of 2026 average 7.1%, in line with the 2022-2024 norm and below every reading from 2016.
ChatGPT reached mass adoption in early 2023. If AI was destroying jobs three and a half years is enough time for the effect to show in the monthly series with a 78-year history. Instead, the series shows ordinary cyclical noise inside its ordinary range.
Individual firms have cut analyst classes and junior coding jobs. But vivid stories from a few visible sectors are a poor sample of the labor market with millions of young workers. Displacement in software and consulting can coexist with absorption in healthcare, construction, and services. The aggregate unemployment rate for young workers displays no downtrend.
Keynes said that in the future', the work week would be 20 to 30 hours rather than 40 hours or more. More vacation time, etc. Current observers are correct when they predict many current jobs will disappear. However, Alfred Marshall, in the late 19th century, observed that hunlan wantd are insatiable. I will mot make projectins a out how AI will reshspe the world because, according to the great Americzn philosopher, Lawrence Berra, predictilns are hard, especially about the future. Where's my jetpack?
High end nannying. The people who can afford it don't want Coriander and Rufus cared for by a Chinese bot. The threat to them, just as to the sommelier, is a shrinking of the market that can bear their price.
One of the struggles I find in my position is that artificial intelligence does tasks well enough, and that people don't expect better output. I wonder what value I bring when other things are "good enough" for people.
I think the question is how messy does a job need to be before AI can't do it.
For eight years in the late 90s/early oughts, I worked as a project engineer for a firm that did outsourced engineering work for food processing and pharmaceutical applications, which I would say is about as messy as it gets. But even then, there were software tools that made some tasks easier and less time consuming. AutoCAD is one example. It made an enormous difference in the time it took to create drawings. Even then, it could create a bill of materials for you. You could also create and store reusable "objects." It didn't eliminate draftsmen, but it reduced the number required.
Now, not only can AutoCAD can generate a drawing automatically using a preconfigured template and a defined list of standardized "objects," it can do much of the design work that conforms to the requirements of various codes.
But neither computer apps nor AI can yet program for automation applications, which cannot be reduced to a standard set of instructions. IMO, that will be the case for some time to come - until cognitive AI advances to a higher plane.
Just yesterday, an acquaintance told me that I no longer need to take notes at meetings of an organization for which I am the secretary, because I can just record it and feed it into a program. I welcome that. Looking forward to learning what program she uses. This seems like a prime AI target job to me--and my arthritic fingers will welcome the rest.
The philosopher, Alan Wilson Watts used to say “we should credit ourselves for the work done by machines on our behalf.” He said this in either the very early 70s or late 60s. I think he was anticipating a robot economy, but I’m not sure he was thinking AI.
I had a messy job: senior automotive regulatory engineer. Negotiation was central to the job. For example, the Pedestrian Safety Enhancement Act of 2010 required the National Highway Traffic Safety Administration to develop a regulation to require electric cars and hybrid cars running in electric mode to produce synthetic noise below 20 mph (the speed above which a vehicle's tire and aerodynamic noises become conspicuous and predominant to nearby pedestrians).
Besides the technology challenges (the technology didn't exist in 2010), this would be a regulation that would affect every car manufacturer in the world. Then there was the worldwide FOMO effect: other countries enacted their own laws to ensure they were seen as proactive and to have a seat at the negotiating table. Some countries, like Germany, had antithetical environmental laws mandating the REDUCTION of vehicle noise. NGOs like the National Federation of the Blind were also involved. Inventing compliance testing machinery and methods was maddening. In the process, I learned to navigate neighborhood sidewalks in Baltimore blindfolded, and wore 3-piece suits to post-Fukushima (no air conditioning!) multi-day meetings in Osaka, Japan.
The final regulation - FMVSS 141 - went into effect twelve years later.... but is still being argued.
I have a question. Do companies use stripped down versions of AI for their service / support chats? Almost every one I have interacted with seems dumb as a post. They don't venture past a "scrpt", have no power of judgement, and seem inadequate at best.
You write that "junior roles are disappearing". If AI is dismantling the bottom rung of the career ladder, the the reduction in jobs should appear in the unemployment rate for 20 to 24 years-olds. It does not. The current reading of 7.2% sits below the 2016 average of 8.4%, a year when nobody worried about AI eating entry-level jobs. The 2025 rise to 9.2% looked like the start of something, but it has fully retraced. The first five months of 2026 average 7.1%, in line with the 2022-2024 norm and below every reading from 2016.
ChatGPT reached mass adoption in early 2023. If AI was destroying jobs three and a half years is enough time for the effect to show in the monthly series with a 78-year history. Instead, the series shows ordinary cyclical noise inside its ordinary range.
Individual firms have cut analyst classes and junior coding jobs. But vivid stories from a few visible sectors are a poor sample of the labor market with millions of young workers. Displacement in software and consulting can coexist with absorption in healthcare, construction, and services. The aggregate unemployment rate for young workers displays no downtrend.
Keynes said that in the future', the work week would be 20 to 30 hours rather than 40 hours or more. More vacation time, etc. Current observers are correct when they predict many current jobs will disappear. However, Alfred Marshall, in the late 19th century, observed that hunlan wantd are insatiable. I will mot make projectins a out how AI will reshspe the world because, according to the great Americzn philosopher, Lawrence Berra, predictilns are hard, especially about the future. Where's my jetpack?
A
High end nannying. The people who can afford it don't want Coriander and Rufus cared for by a Chinese bot. The threat to them, just as to the sommelier, is a shrinking of the market that can bear their price.
One of the struggles I find in my position is that artificial intelligence does tasks well enough, and that people don't expect better output. I wonder what value I bring when other things are "good enough" for people.
An excellent article!
I think the question is how messy does a job need to be before AI can't do it.
For eight years in the late 90s/early oughts, I worked as a project engineer for a firm that did outsourced engineering work for food processing and pharmaceutical applications, which I would say is about as messy as it gets. But even then, there were software tools that made some tasks easier and less time consuming. AutoCAD is one example. It made an enormous difference in the time it took to create drawings. Even then, it could create a bill of materials for you. You could also create and store reusable "objects." It didn't eliminate draftsmen, but it reduced the number required.
Now, not only can AutoCAD can generate a drawing automatically using a preconfigured template and a defined list of standardized "objects," it can do much of the design work that conforms to the requirements of various codes.
But neither computer apps nor AI can yet program for automation applications, which cannot be reduced to a standard set of instructions. IMO, that will be the case for some time to come - until cognitive AI advances to a higher plane.
Just yesterday, an acquaintance told me that I no longer need to take notes at meetings of an organization for which I am the secretary, because I can just record it and feed it into a program. I welcome that. Looking forward to learning what program she uses. This seems like a prime AI target job to me--and my arthritic fingers will welcome the rest.
The philosopher, Alan Wilson Watts used to say “we should credit ourselves for the work done by machines on our behalf.” He said this in either the very early 70s or late 60s. I think he was anticipating a robot economy, but I’m not sure he was thinking AI.
Really fascinating, messy stuff!