From assistance to agency: what happens when AI begins to perform the work of engineering?
Volume II · written from October 2026, two years after Volume I
The original Generative Engineering study asked what would happen when artificial intelligence joined the engineering team. Two years later, the question has become less comfortable—and more interesting. In mathematics, laboratories, software, factories, and robotics, AI systems are beginning to do more than propose. They can act, test, revise, and sometimes discover. Not everywhere. Not reliably. Not without people. But sufficiently often that the nature of engineering itself deserves a second examination.
Start here · The Second Act
From assistance to agency—and the changing object of engineering.
Go lived the second act first: the machine that could not be defeated made human players better. What that means for engineers, now that thought is abundant but not cheap, and judgment is the scarce resource.
Read the essayFour essays
01 · Essay
Proof may become abundant. Understanding may not.
In 1957 a correct proof showed that a sphere can be turned inside out, and nobody could see it; a blind mathematician was among the first who did. What happens when such proofs arrive by the hundred, and why correctness, significance and understanding come apart.
It ends on: “If proofs become cheap, does mathematical taste become the scarce resource?”
Returns to Mathematics and AI, Trail 1: Engineering Optimization Unleashed and Trail 3: AI as Your Problem-Solving Ally
Read essay 01
02 · Essay
From collaboration to delegated action.
An assistant you text now books, pays and calls in your name. The French Civil Code of 1804 already knew what to ask of such an agent: what it may do on its own, what needs an express mandate, how it renders account, and how it is revoked.
It ends on: “If an agent can act without asking, what should force it to ask?”
Returns to Trail 3: AI as Your Problem-Solving Ally, Engineering management and The Engineer's Moment
Read essay 02
03 · Essay
From simulation to experiment to action.
Steve Rogers lowers the flag by pulling the pin instead of climbing the pole. What changes when machines can redesign the problem, run the next experiment and act on the result: a mirror with a purpose, a laboratory that answers back, and the right to stop the loop.
It ends on: “The machine can help us find many more pins. Who decides which ones we are ready to pull?”
Returns to Trail 2: Smart Digital Twins, Trail 1: Engineering Optimization Unleashed, Industrial engineering and systems and Robotics and Control Systems
Read essay 03
04 · Essay
The apprenticeship paradox of automated expertise.
I can build an interface in TypeScript; I do not know TypeScript. What AI does to the person inside the suit: help that teaches and help that replaces, the colleagues who stay in your head, the tutor in every pocket, the missing bottom rung, and who pays for the next engineer.
It ends on: “What kind of working world will produce the engineer who decides what the next suit is for?”
Returns to Education and training, Engineering management and The Engineer's Moment
Read essay 04Keeping score
In early 2025 we made four bets about how AI would change engineering. I keep the score in public, against 28 sources from my essays, two or three per question, and I say how I chose them. So far, one went further than we said and three were partly right; none has been contradicted yet.
Scored by the person who placed the bets, so watch my hands: every source is linked, what goes against us comes first, and a log shows every time the score moved.
Explore the original study
The background, four trails and their twelve trends, two deep dives and the epilogue.
10 engineering fields, each a chapter with its sources.
16 conversations, one for every chapter and field.
How the study was made, with its method and its authors.