All he made was some dinky algorithm. Google Bard could do that in three minutes flat smh.

Thx for sharing this . Really hope people read it.

See, Dijkstra was talking about people trying to create programs in natural language. He didn’t say not to use your natural language to hire someone else to make a formal program. This is people using natural language to hire an LLM to make a formal program, and asking LLMs is like asking people, so it’s Dijkstra-approved.

My managers are supposed to be skilled?

Rust & cargo do more than just compile. For example, it basically has buit-in ccache.

It is also easier to split large libraries into multiple crates, though an average project still uses more libraries than an equivalent C project. I wouldn’t be surprised if the “AI” also pulled in more libraries than needed, or has unnecessary library features enabled. I’m pretty sure that a cargo plugin for pruning unused libraries was featured on the rust blog, as a featured third-party plugin for a cargo release.

Amen.

Safety critical (aerospace, medical, precious few other) industries have regulated quality, with moderate success. It’s far from perfect, farther from ideal, but it is providing some additional resource and schedule allocation to do the things that need doing to ensure the systems don’t screw up too badly, too often.

One of my junior devs was having trouble with a bug in an internally developed tool, apparently for weeks before I saw her struggling with it over her shoulder - it was a 5 minute fix, I hope I made it clear to her: speak up when something’s wrong - this 5 minute fix has cost you many hours already because you never told me you were having a problem.

In the 80s the hand written assembly was more reliable and performant than the C, at least on many of the compilers.

Even in 1990, I tried to launch a serious project in C++ on the IBM-PC, and the best available compiler was too buggy to use. It did fine for little demo apps, but by the time you wrote code for 2 weeks, you started hitting bugs - not in your code but in the compiler output… we had to fall back to C for the project. Even later, around 1994, we had two C compilers for 6811 work and one of them was garbage, I could hand write the assembly better and faster without even trying hard. The other one was pretty good, and by the late 1990s I stopped looking at C/C++ compilers’ assembly output because it was consistently better than I would write by hand.

is there any other kind?

The SOTA changes every couple weeks, but Claude’s been very dominant for a while, yeah. There’s currently a lot of hype around GPT-5.4, but even then there’s a caveat that Claude is still better at UI.

I just personally find Cursor to be pretty buggy. But I think the Replit mention is more of a tell that someone vibe codes but doesn’t actually code. It’s been advertised to people as a way to build end to end apps without any coding experience. And to be fair, they’ve done a good job of building on the past decade of work in the Typescript community to make an entire app end to end type safe and therefore checkable by the compiler. Convex has done something similar in a way that I prefer and in my experience LLMs are very good at working in Convex projects as well.

Really at the end of the day I was just being pithy. Kind of poking fun at how much of a moving target SOTA is.

I am vastly prefering copilot over claude, using sonnet 4.5~4.6 for most tasks and then pulling out opus as “the big guns” for tougher stuff sonnet cant handle easy

Copilot is only costing me ~$28 a month, which gets me 1500 premium requests per month

If you set up your flows well, 1 premium request is an entire session, so Im only paying like 2.4 cents for 20 minutes of work

ate $15-20 Claude credits in less than ten minutes.

Lay off of MAX mode.

Also, if you’re paying API rates, look into the subscription options - I can’t burn the $200 subscription plan down much below 50% without pushing prompts into Claude every waking hour (unless I turn on MAX mode). At API rates? I can burn $50 in a few hours.

do what I haven’t ever done: decline the whole PRs as they contain a lot of architectural changes that complexify the system in order to achieve the goal.

If you’re accepting the first thing the agent gives you, you’re almost certainly “doing it wrong” - gate it before it goes down a bad rabbithole and redirect it, in writing, in architecture documents (which it can draft for you, and correct based on your guidance) - and when it ignores those architecture documents, which it will do when things get big and complex, break the architecture documents down into smaller chunks that apply to the various tasks at hand - yes, it can do this breakdown for you too and that’s another opprotunity for you to guide the process. I try to frame the output I get from AI in my mind as: usually about 80% correct / useful, and it’s my job to identify that other 20% (which, in reality, is getting a lot smaller lately), and beef up the specifications and descriptions of the job until it can get everything to an acceptable state.

I don’t trust it and have to recheck every single scenario. I trust junior engineer in our team more than I trust this instrument.

That would depend entirely on which junior engineer your are talking about, for me. I don’t trust Claude, either. But for the most part I have Claude check itself, at an appropriately granular level. If you’ve got more than 2000 lines of Claude’s code that doesn’t have good visibility into what its doing, why its doing it, and what the outputs should look like… you’re trusting it too much. But it can write that documentation and testing for you, you just have to review it - at an appropriate level. If you’re trying to do it line by line of code for a big project, maybe you should still be writing it yourself instead.

It’s not a “narrative”; it’s their experience. I don’t have the same experience, but do have experience of myself and colleagues using LLM agents effectively and doing more work reviewing their output than writing lines of code. Some colleagues are pretty much AI boosters, but most are very aware of its limitations.

nah such narratives are mostly pushed by Ai companies

Someone’s personal experience is an AI company narrative now?

Opus 4.6 is genuinely pretty decent at programming now if you give it a good backbone to build off of.

Soup from a Stone.

To an extent, yes. The more “broth base” I feed Claude, the better it does. If I just vaguely describe a program, I get a vague implementation of my description. If I have a big, feature rich example (or better, examples) of what I want the program to do, Claude can iterate until the program it make’s output actually matches the examples.

I’m not sure if I’m reading your intent correctly or not, but the AI agents actually excel at “puddle jumper” tasks. Stupid stuff that you could write a one-off script for, but damn that’s a lot of hassle. This afternoon a colleague and I were putting together a powerpoint slide deck based on a folder full of disorganized garbage. Claude digested the garbage and wrote a python script that wrote the .pdf and I’d swear it took almost as long to open Office 365 Power Point as it did for Claude to write the script that wrote the 7 slide .ppt file.

roadmap definition, sprint planning, unit tests, regular independent code reviews, etc. etc. etc. that those of us who “knew what we were doing” mostly looked down on as the waste of time that it was

You sound insane.