I can verify that from AI coding: I get the AI agent to write a commit for me, looks good and tests parse. But much later I discover a lot of issues, I simply was too lazy to find in my own review.

You cant do that: Agent based LLM code generation actually incorporate the result of compile and test. The same quality checkers are as if not even more needed when the code is node by a LLM which is as error prone as human.

I askrd Gemini to set up a unittest for a state machine I had coded in Python Instead of calling the methods changing the state and checking the resulting state, it made code changing the state directly and tested that. The production code was not tested at all. Maybe my fault for not making internal attributrs hidden, but still...

If someone downloads it from Afghanistan and Taliban outlaws it, who cares? Why care about Brazil? Maybe just write, which juristiction does apply to the project, and ignore every other stupid law out there.

Tools scanning your git repo for secrets exists, but you have to recognice the problem before running them. But the problem is quite common, because it is a little but more complicated to add an indirection to a secrets vault.

In the US you mostly use rails for slow moving cargo, and perspn transport is by car and plane, while in Europe the rails are occupied by faster going passenger trains, and cargo is by truck. Our local container terminal is serviced by trucks driving through town although there is a rail from old time going down to the harbor.

There a still a few old IDEs you can script with TCL...

How many times does review actually catch bugs? Why should review wotk better for AI generated code than human written? For both you need some rigorious, i.e. non-AI, systems ti catch bugs: tests and static code analysers. If you have too many bugs, up your testing. Use the tests to safely refactor provlematic code. Review is at best a way to align developers and train newbees, not Q.

As long as investors believe in it, it keeps going. Just as with any other bubble.

A few years back every child in the public schools in our town got a Chromebook. My now 10 year old got one in 2nd grade, but in 3rd grade everyone had to hand them in again. Also, somewritten exams are without computers in high school again now. In short: computers and screens are bad for learning (in the lower classes at least).

Yes, but you need a quantum oracle to coherently calculate if a state x is in S or not.

Many banks happily replace IBM with Microsoft. They don't care about vendor lock-in.

AI can translate code the same as humans doing a lot of mistakes, just faster. You need a torough test to verify the translation no matter what. Unfortunately, making automatic tests was not Cobol's strong side. It wasn't that common back then, and even now a lot of software projects lack even any automatic test suites.

The Norse, who settled on south west coast, died out in Greenland, when it got colder and their farming and coastal fishing didnt work anymore. Contact between Norway and Greenland was lost around 1400. The Danish-Norwedian king tried to reestablish contact in the 18th century expecting to find christians, but still catholics, up there, but found Inuit, who had come in from the north. Denmark lost Norway to Sweden after the Napolian wars, but kept Greenland, Faroe Islands and Iceland.

"Rocker science" is really "rocket engineering" - which is hard by itself due to the extremes of it, but not science as such. There can be pockets of science, say, material science at cryogenic temperatures, in there, though - but 60 years ago, when the technology was invented.