This is exactly why I advocate people write code, even in a world where agents code “better”
It’s not enough to read code. You don’t internalize it. You need to experience it painfully :)
Also agents frequently defensively wrap old code instead of questioning whether legacy paths should exist. You get this nested doll layering effect of extremely large amounts of defensive code.
Seems the smart thing to do is not assume an agent will do the right thing. But to create the scaffold / harness that enforces constraints to steer them towards a good result.
Then you can swap out the really smart model for maybe something cheaper.
Does this model do better ignoring side conversations? That's the biggest hindrance to using ChatGPT's carplay feature is someone will say something, stopping ChatGPT from speaking or taking it in a different direction.
I also think we’ll approach a point where increasing intelligence is not really going to suddenly improve most work tasks. I bet that’s already happened actually.
We’re oohing and aahing about models, when the ones a few versions ago did a good enough job for most of the dumb coding, etc we do
At best the conclusions are a best case scenario for degraded code quality. That things remain functionally OK with more costly token usage to get work done.
I can totally see doing this incrementally, but this seems extremely risky to do for the entire codebase in one shot on anything in production. Especially if you don’t have really thoughtful e2e tests of the whole system.
Yes this is one reason IMO I think of AI code as instant legacy code.
You take on a lot of tech debt. Then you need to do the same work you would do with any legacy app: finding where the brittle points are, what needs better testing, which leads to breaking apart the big ball of mud into cleaner components.
All it takes is some brittleness to need to take apart the app code and understand it. AI fixes one thing, breaks something completely unrelated.
And usually it’s full of lots of spaghetti code slop. Then I need to find ways to modularize it, make it testable, and at least at some level clean it up.
This is why I say AI sw eng is basically just working with legacy code. It has always turned into unraveling and refactoring the ball of mud to be sane for me and agents to continue working with.
More and more though I just go slower. Write much of the code myself. And setup good validation for the specific parts I trust an agent to work autonomously. I try to expand the surface of things that can be done well autonomously without losing my own grip on the code.
The study looks at junior developers unfamiliar with what they're implementing. Post-hoc they broke down the AI group and grouped some into tutoring->hand coding. Another group was AI maximalist. We should keep in mind the comparison of these groups is very low n. Tutoring + hand coding seemed to have the best speed + understanding. This was all conducted in Jan 2026.
Where I'd push back drawing too many conclusions from this study: arguably most successful AI usage is senior developers that know the programming environment they're working in. Know how far to trust the AI. And carefully review / understand outputs.
Nevertheless, the study's still interesting, and I wish they'd replicate with a much higher n per group. Junior developers (undergrads?) are a more abundant group and not particularly specialized yet. They've also spent years hand-coding at University, but probably could adapt to AI tooling pretty easily.
Anthropic thinks they’re managing safety. But they come across as just elitist. They appear to say only they should get to make safety decisions about their creations
But the conflict of interest is plain to outsiders. And they seem to lack skill in working with public institutions to create consistent safety legislation. Any regulations they advocate for just seem like rent seeking.
We just need people with expertise in government. Not Anthropic/OpenAI making these dumb decisions.
I still oscillate between "I'm totally cooked, I have no role here, the AI does everything" to "WTF why is this LLM so stupid today, WTF is it doing? This is garbage?"
A lot of that is because in the former case (AI does everything) I wasn't paying enough attention.
The real problem in all this is lack of predictability. The White House is just making it up as it goes along. Investors, customers don’t know what the process is and can’t plan.
In the end, we need actual laws that tell the market what kinds of models get paused / analyzed, how long that pause can be, etc.
Otherwise there’s no standard and it will be easily abused and prevent investment in US AI companies.
The real problem is the White House just making up the rules as it goes. No laws. No predictably for the markets.
A week or so pause from seemingly legitimate cyber security concerns isn’t cause for panic. But it should be backed by laws that describe what that process should be. That would put the market at ease
Like equivalent to what your developer salary would be? Less? More because you’re a CEO now?
I basically have minimum $ amount I would accept for “developer job I absolutely love” that I know sustains my family comfortably without much extra fun or savings. Is that a good bar?
http://softwaredoug.com