A team of human specialists will be better in the long run.

For the 0.1% of the population who can afford to assemble a team of human specialists in a timely fashion when they need them…

For the rest of us, cyborg M.D.s with brain augmentation (they called this a PubMed search in the 1990s, it’s faster and more effective today) are better than the un-augmented M.D.s we can afford.

According to the Anthropic blog, they terminated the Houthis’ accounts - so now they’ll use ChatGPT more carefully…

The thing about medical is that it’s so big - no one expert can cover even a small fraction of “medical”. So, unless you have a whole team of specialists there on-hand, you just might benefit from having some “automated experts” available when you need the pattern recognition they can offer.

People who rely too heavily on anything get brain atrophy with net negative consequence.

Those same tools can be used for learning and exploration and strengthening skills - all a matter of how they’re used.

Not all metrics are good. LOC is about as bad as they get.

Bugs per merge is a much better one.

Last i heard was that developers FELT

Several of those studies were published on early 2025 data, LLM agents have been rapidly improving since then. In early 2025 I would agree with the sentiment - it was kind of equivocal for overall productiveness. By early 2026 I’d say it was a much clearer advantage, both in doing the same kind of work faster, but more importantly: in doing better work at the same speed or even a little faster.

the code that made literally everything until a few years ago? That was crashing?

Quite a bit more than it should.

If Vibe coding means people who don’t know how to code creating code… yeah, that’s kind of like giving people who don’t know how to paint an airless gun and 5 gallon bucket and telling 'em to go for it on this building here…

What happened to those developments that started in the 70s where you have automatic formal proofs that a piece of code does what it is supposed to do?

The stochastic parrots we have today are pretty good at doing that kind of tedious, extensive, franky mind numbingly boring stuff. They’re also fast enough that you can “afford” to take the time / effort required to do those kinds of proofs where they are warranted - it makes them “a good idea” in a lot more places than they used to be when they were so much more expensive to perform.

Non AI code has been crashing down for decades, controlled chaos, continuous improvement. We have lots of tools to put metrics on that, and AI assisted code is fairly well crushing those metrics. It’s not 100% self sufficient code creation Nirvana, but it might be fairly compared to the Cotton Gin in terms of making the field workers more productive in the delivery of market ready code/cotton.

I lived in a city that was doing a lot of lane count reduction, installation of traffic circles, etc. Results were… mixed. It definitely was achieving “aesthetic improvement” with more trees and green features along what used to be bare asphalt and concrete corridors - sometimes the traffic did “flow better” on the lane count reduced roads, but I think a lot of that was attributable to significant volume diverting to alternate routes. Then they had their “disaster area” situations that only got worse no matter what they did - overdevelopment underserved is a formidable problem.

The main thing I find with the LLMs is that they’re very fast and capable and limited. You give them a task of low complexity, they can execute it with superhuman speed, and if there wasn’t hidden complexity in it, they do well.

If the task is too complex (beyond their context window’s capacity), they start to “hallucinate” - make up things that sound reasonable to their training - to cover gaps in “what to do next” because they’ve forgotten some of the complex instructions. Putting the instructions “in writing” in a reference document that the agent doesn’t edit freely can help a lot, but eventually those documents get too complex as well and again it starts “winging it” rather than endlessly re-reading the instructions before making every next move.

The models’ training is getting better, doing “the right thing” more often by default - without requiring as much explicit guardrailing. They’re also getting better at paying attention to requirements documents, but they’re far from perfect. As are humans, but humans don’t always move as fast as LLM agents, so LLM agents can make big messes faster…

Agentic AI has its place, in a tightly controlled sandbox.

One of the anecdotes was about an agent which wiped out the production database, and all backups. While you can lay blame on the agent (AI or human) who made the mistakes, the real blame in that situation is with the system architect (likely there was no named system architect, but that doesn’t absolve whoever was acting as de-facto architect) who placed the backups in a hot, online accessible and erasable configuration. No agent, AI or human, should be able to wipe out all the backups with their normal access - ideally there are physical doors and keys involved - in multiple locations if the data is of any value.

You know, it’s really about poor… black is shorthand for poor, so it looks like racism, but underneath it all, they’ll abuse poor people just like they abuse black people.

If they’d make driving less convenient and transit more convenient, the number of cars would decrease significantly.

Of course you’re talking about billions of dollars in construction projects, is it you in the construction industry, or a family member? It took 100 years of infrastructure build to get here, it’s going to take quite a few to change course.

All depends on how the town was built.

Coral Gables has laws against carrying “burglar tools” - well, a credit card or even a state issued ID card can be used to open a locked door, so there’s a handy “burglar tool” charge just as soon as the cop asks to see your ID.

When the cops in our town were ineffective at keeping the teenagers from loitering in the parking lot of the 24 hour drug store, the retiree busy-bodies got the city council to make the drugstore close at 9pm.

I had a friend who had an apartment in Hamburg, roughly 600m from an S-Bahn station. within 50m of that station was the neighborhood grocery. You walk up to take the train to wherever, and on your way home you stop off in the Biergarten first, have a few, then stop in the grocery before walking home - downhill no less for the arms full of packages side of the trip. Not a bad layout, but about half the residents still had cars parked all over the streets which they used more or less daily. So, in some ways, the US arrangement just doesn’t bother with the under-utilized trains, but it also eliminates them as an option which really sucks when you want to go to the airport, or take an inter-city train somewhere…

More cars per hour is less cars stuck in traffic on the roads… the number of cars is a given, getting them off the road ASAP is the goal.