SWE's have to guide the AI, look at all those AI PR's, many of them make very bad assumptions.
As a result I decided to start using those LLMs myself, to see what they could do and how quickly they would make me obsolete. Two months later, I feel a lot less worried. Yeah they're really good for some things and they do save me a lot of time in some cases, but I wouldn't trust them to develop a complex application from the ground up without oversight. If it's a one-off prototype thing then maybe, but for a serious long-lived project it's going to be utter chaos and confusion.
One of the reasons the bots are so good at pure coding is that code is designed to be machine-friendly by definition, and the feedback loops are extremely short and cheap. Oh you forgot to add an #include? Read the compiler's error message and fix it. Takes 2s. I suspect that's also true for the Mathematical proof work to some extent, your Lean either checks or it doesn't. As soon as problems become more organic and less immediately testable, you can see models losing their footing. When there isn't a clear test to decide between A and B, the models easily panic, loop, overthink, underthink, miss alternatives etc...
Of course it may change. Maybe next year Opus 15.5 will truly outrank me. But there's still a long way to go IMO.
So many executives have turned all the safeties off and given over control of their business to a probabilistic system and ignored every engineer who said its dangerous and risky. The engineers will be the ones held accountable though, because since AI is "so effective" it must be operator error
Second point should be obvious to anyone actually building something even slightly new/complex.
2) I think u might be right about brute force - but probably some methods/tools will come along first - and I don't think they will solve the problem though.
I think SWE is actually very safe.
We're going to need fewer 'JS coders' and technically we could 'save' a few SWE jobs - but - given the $ being made, businesses will want to expand, and so 'Jevon's Paradox' and I think SWEs will have jobs and probably even the JS jocks as well.
Ofc unless you call some data modelling “architect” but I’m sure AI is fine for that. Guess how I know.
This does not match my experience. At work I have AI write architecture docs for complex features based on requirements that the product team collects. When staff engineers review those docs, in the overwhelming majority of cases the only issues they can find can be characterized as nits or bike-shedding. In contrast, AI can find problems and even serious gaps in architectures designed by the most seasoned engineers on the team (because those engineers have been doing their best to resist using AI).
So I think the whole "AI can write good code but it cannot design architecture" is just cope. It's where the goal posts have been moved to after it became impossible to deny that AI can actually write good code. Next up it will be "AI can write good code and design good architecture, but it has no taste". Indeed we're seeing various of that already.
You still need someone at the wheel with a strong sense of where the architecture should go. If you just let AI drive, it goes in circles forever.
Just had this experience at work. I tasked a mid-level engineer to architect a brand new complex system with AI assistance and he spent a month burning tokens trying to get everything specified. Between AI going down irrelevant rabbit holes and him not being assertive enough over the AI, it didn't work. After a month, there were still tons of open questions, undecided workflows, etc. I wanted to try this as an experiment, and to give him the opportunity, but it didn't work. And this person is not at all incompetent, just doesn't have enough experience in big-picture system design.
I reassigned a very senior engineer to take over the project on Monday and in a few days (writing this on a Thursday night) the design is done and implementation is well under way.
Ai is not a magic silver bullet. It's a wild horse and you need a master horseman to tame it.
AI can write syntactically perfect code but barely has a sense of architecture.
It needs tons of 'guide wires' to keep it from going off track.
Unless you're building basic SaaS, it's still a 'research and coding assistant'.
AI has never produced non-messy architecture that I did not have to go in, re-think and guide a re-write.