I have been doing daily agentic coding myself since last November, but lately Im having concerns. We are on the verge of becoming Warhammer 40k Tech-Priests: chanting sacred prompts to appease the machine spirit, praying for a good result. The foundational knowledge is lost, meaning that even if we do look at the code, we won't understand it anyway.
but this is similar to a lot of other technical fields that have width and depth. People would not understand assembly any more and blindly trust the output of the compiler. People now can only design a part of a CPU, and not the entirety. Chip manufacturing is so complicated that there's probably no single person who could understand the entire process, from end to end.
Yet, we all trust these things.
It doesn't always land on the right person but it's usually a motivating factor for people to make an effort.
This was OK as long as you understood enough Assembly to debug it. Even though I haven't written much Assembly directly, I can still debug it when there are compiler bugs, which does ocasionally happen.
If you lose the ability to both write and debug code then things get much worse. If no one at a company can do that, then you truly are at the mercy of the LLMs.
As for hardware bugs, those are only possibly to identify (by excluding any other possibility) and work-around, since most engineers don't have the luxury of being able to change the hardware.
I'm wondering if some of us will become consultants to companies that lost the ability to handle code. When I was in university the going meme was that old COBOL programmers were making a lot of money in consulting jobs, maybe this will be the new version of that.
There are YouTube videos of a guy doing chip fabrication in his clean room shed. Just as I understand the assembly output of my compiler (… and no, I don't work in a low level language), he understands chip fabrication end to end.
CS is not that wide, not yet. No, one cannot know every facet of the field, but one can know enough about the facets one is in contact with that they're not "here be dragons" on the map. All the way down to the hardware.
The fact this is "a guy", and probably there are very few people in the world who would be able to execute such a thing, probably should tell you how difficult that actually is. You're actually unintentionally making parent's point stronger, as obviously the production processes used for market products is much, much more complicated than what a guy in a shed does.
> Just as I understand the assembly output of my compiler
Fair enough, this is probably way more common than "person who understands the entire chip manufacturing process from beginning to end", but it's still something that less and less people know about, in a room of 100 average developers, probably few of them could realistically claim to understand all the assembly output of their compiler, if even they're familiar enough with that their language outputs assembly in the first place.
The range of skills in developers is enourmous, you have Klaus working as IT+web dev+database+admin in a tiny 2 person web shop, and on the other hand you have Anders who is a ML expert optimizing CUDA kernels, and everything in-between. That all (or even most) of them know the far end of what they currently know is wishful thinking really, Klaus doesn't need to know about what the NVCC compiler outputs, nor do Anders need to know about the specific constraints of mapping packaged products to order shipments, that's all fine.
You could just as easily point out that subsistence farmers exist. That’s great, but we still need industrial agriculture to feed everyone. No single human fully understands that domain either… even ignoring the chemistry, biology and meteorology, those farms use digitally controlled tools! That means all of computer science is embedded in farming.
It’s great to pursue broad knowledge. It’s important to understand your dependencies and respect the abstractions you build on. It’s pure hubris to think an individual auteur can challenge a multi-billion person civilization.
We trust them - in large part - because we know that professionals took the care and responsibility to build the foundations correctly.
And if the foundations failed in some way, those same professionals would suffer a hit to their reputation.
Yes! Because of two words. Consistency and Reliability.
All the physics equations, we trust them not because they are pretty, but because they have been consistent with experiment. Simplified models of transistors are also consistent within a temperature range. Boolean Logic is also consistent, the Turing Machine is consistent,…, the electronics inside a chip is reliable. The instruction set is reliable, etc,…
> Chip manufacturing is so complicated that there's probably no single person who could understand the entire process, from end to end
I strongly doubt that. The actual theoretical knowledge is taught at school and you can take an iron and some electronic elements and build some 8bit computer. Modern chip manufacturing is a matter of scale. It’s cumbersome to design with logic gates, so we invented hardware design languages (just like it’s cumbersome to deal with opcodes) and they are reliable. And instead of PCB, we just use a silicium wafer.
Not yet. The world still pretty much runs on systems not written by LLMs. And humans wrote all of that. LLMs runs on OSes and hardware created by "not LLMS".
And seen the "quality" of the output of LLMs (and I pay three AI subscriptions), I don't think we're getting rid of human-produced code, OSes, hardware, anytime soon.
I tend to think it's just different personality types. There are a lot of different type in engineering. There's a significant set of people that really love to just be given detailed, well described set of work and crank it out, they don't really want to be involved in much outside of just implementing the plan. That is not all engineers but certainly a chunk of them. I could see that group feeling a bit sidelined by AI. But that's not me, so I am just speculating.
You can’t not claude code if everyone else is pushing to main (or opening prs which get ai reviewed), and once your days become mostly prompting and ai code review, you start remembering the days (mere months ago) where work did not feel or look like this at all.
The condom image is both an exaggeration, but is also almost a too-accurate depiction of what work has become - very rapidly - for many people.
I’m sure people were also very excited to get their first cars after years of riding on horses. It doesn’t mean new kinds of pollution (and accidents) are not happening, or that we don’t miss horse riding haha
Cheers
He wasn't wrong, and his role was still valuable to the company.
Maybe that's you. In my case, it's going better than ever.