Giving a precise description of what the computer is supposed to do is exactly what programming is.

The more specific your requirements the closer you get to natural language not being useful anymore.

Yes, there are still many areas where skilled humans are faster than AI (meaning faster coding yourself, than providing so much context and guidance that the AI can do it on its "own").

But in general the statement is really not true anymore, generic projects/problems have a pretty good chance that the AI can one shot a working solution from a lazily typed vague prompt.

Maybe a failure to automate?

The volume of people successfully adopting agentic engineering practices suggests this stuff isn't rocket science, but it is a learned skill and takes setup.

A year later into heavy AI coding, my experience is what you're describing should aid in being able to run 5+ agents simultaneously on a project because you know what you're doing, you set it up right, and you know how to tell agents to leverage that properly.

Yeah it’s when you go off the happy path that it gets difficult. Like there’s a weird behaviour in your vibe-coded app that you don’t quite know how to describe succinctly and you end up in some back-and-forth.

But man AI is phenomenal for getting stuff out of your head and working quick.

That doesn't matter. The statement wasn't "faster than AI right now", it was "will always be faster than AI". And that's just nonsense.

Current AI systems are extremely serial, in that very little of the inherent parallelism of the problem is utilized. Current-gen AI systems run at most a few hundreds of thousands of operations in parallel, while for frontier models, billions of operations could be run in parallel. Or in other words, what currently takes AI 8 hours will take it barely long enough for you to perceive the delay after you release the enter key.

For a demo, play around with https://chatjimmy.ai/ , the AI chatbot of Taalas, where they etched the model into silicon in a distributed way, instead of saving it in RAM and sucking it to execution units by a straw. It's a 8B parameter model, so it's unsuitable for complex problems, but the techniques used for it will work for larger models too, and they are working to get there.

And even Taalas is very far from the limits. Modern better quality LLM chatbots operate at ~40 tokens per second. The Taalas chatbot operates at 17000 tokens/s. If you took full advantage of parallelism, you should be able to have a latency of low hundreds of clock cycles per token, or single request throughput of tens of millions of tokens per second. (With a fully pipelined model able to serve one token per clock cycle, from low hundreds of requests.) Why doesn't everyone do it like that right now? Because to do this, you need to etch your model into silicon, which on modern leading edge manufacturing is a very involved process that costs hundreds of millions+ in development and mask costs (we are not talking about single chips here, you can barely fit that 8B model into one), and will take around a year. So long as the models keep improving so much that a year-old model is considered too old to pay back the capital costs, the investment is not justified. But when it will be done, it will not just make AI faster, it will also make it much more energy-efficient per token. Most of the energy costs are caused by moving data around and loading/storing it in memory.

And I want to stress that none of the above is dependent on any kind of new developments or inventions. We know how to do it, it's held back only by the pace of model improvement and economics. When models reach a state of truly "good enough", it will happen. It feels perverse to me that people are treating this situation as "there was a per-AI period that worked like X, now we are in a post-AI period and we have figured out that it will work like Y". No. We are at the very bottom of a very steep curve, and everything will be very different when it's over.

I don't believe this. Either you're lying, or you just haven't caught on with how to use Agentic AI.

Everything I do to interact with my computer is through an agent now.

Care to explain which particular intimate knowledge allowed you in the last 6-9 months to be faster than AI in certain area?

Honestly, I'm still faster than AI cooking scrambled eggs, but definitely not faster than neither AI (or compiler) in translating stuff into code.

> Some of us do actually have intimate knowledge in certain areas where guidance of an AI takes longer than doing it yourself.

You speak as if AI development is frozen, and you ignore the poster's point:

> that gap will only increase as LLMs get more intelligent

This has been my experience thus far. Yes, a complete prototype can be made, but.. you don't really know until you read the code and test it. Just yesterday, small things came up in terms of Qt screen focus that wouldn't have come up otherwise save for initial testing.

I think, and I recognize it is mostly against the 'agentic' push, I will stick with slow iteration.

My experience is the opposite: AI takes too many edge cases into account and guard against even the most unlikely thing. The upside is that it often handles edge cases that I either didn't think about or was too lazy to implement.

I can with full confidence say that the code AI writes is more robust and safe than if I would have done it myself. The code definitely becomes more bloated though.

It also loves to add edge case handling where it's not needed and in poorly chosen places

Which is still false and not serious. It's one of the dumbest rationalizations I've seen. AI has many flaws but pretending that it's useless because of that is not it.

Also employees who work slower than you (and spend most of their time not actually working).

Especially if you use auto-complete AI, ironically. You type a few characters, the line fills out in less than a second, as opposed to a reasoning model that takes maybe a second per 2-3 lines it writes out.

Explaining your idea of intelligent would have been a better comment than name calling and shallow dismissal.

They very clearly weren't talking about nerds in general but rather nerds who care about software.

But business people always cared only about thr result. My PM (who speaks like a salesman) only cares about the results. My “head of” same. My ceo same. The only ones who ever cared about the process and quality were us the engineers… if we don’t have that care, well, to hell with everything

> enjoy the process

This means different things to different people, lot of people enjoy the process of engineering solutions with LLM agents, build out tailored skilled, custom approaches that make up their own flavour "agentic" workflow. There are also people who find joy in Javascript that other people cannot understand why. And other people again love system languages or even tinkering with assembly etc.

What I wanted to say is that LLM use does not automatically mean people just want to get results faster, there are still nerds enjoying the process of working with these new tools.

I am not really sure. I wrote some scripts that aggregated data from several APIs with an LLM and the LLM had the foresight to create a caching layer for the API responses as it properly inferred that I would need the results over and over again as well as using asyncio to accelerate fetch speed. This would have been a v2 or v3 and it one-shotted it perfectly.

The benchmark is "do they do it for fun", i.e. personal projects.

But the real trick isn't "number of personal projects", but how weird they are. There's no "rational" reason to do them, they don't increase the person's marketability / hireability. They are done purely for intrinsic reasons.

(On reflection, this also seems to be a pretty robust predictor of autism. :)

>in software we accept the worse of the two

and the whole world suffers for it.

"I've posited for a while now" and you post the most lukewarm and outdated take like it's an enlightenment. I've been coding for 20 years and can very well do everything the AI does, and so can all devs I know. We use it because it amplifies us, not because we couldn't otherwise. You've chosen a very ridiculous hill to die on.

Is your argument that there is no imaginable situation where someone who was competent at software development could find use for a semi-automated tool for writing software?

That would imply that either the person in question has infinite time, or has access to all software that could ever be of utility to them, which seems unlikely.

Initially I wanted to write more but I can boil it down to taste and context mismatch. By that I mean some people see LLM output as tasteless or kitsch (which I ascribe to generally) and another set of people (though sometimes overlapping more often than not) hold disdain or at the very least look funny at heavy LLM users like gym-goers would look at someone in the middle of the gym loudly suggesting using a dolly or forklift instead of barbell training.

So yeah, I guess the value of doodles has shot up simply because of optics.

Somewhere else in this comment section someone tried to broaden the definition of nerd so much so that pretty much anybody who is a consummate professional is also a nerd. The hill I will die on is that people don't actually dislike all this new AI stuff but more so the attitude of people heavily invested in it.

And to add another data point regarding your hill my drawing/painting moment was NLP stuff. Now if I want to do (rudimentary) sentiment analysis or keyword extraction I can lean on a local LLM. Yet I don't go around yelling Snowball (I think?) is obsolete.

While you are dying on a hill, with the help of LLMs, I'm shipping quality software and features to my customers at a pace I haven't been able to before. And no, not some nextjs slop. If you are letting your LLM look at StackOverflow, you are doing it wrong - it needs to be grounding in your stacks official docs and any other style/rules you prefer wired with other tooling like linting/formatting, duplication checking, etc. And yes, you have to constantly monitor the output and review every line of code - but it's still faster and if managed correctly, produces better code and (this is the hill I will die on) better test suites and documentation than I would have written.

I think there's some ambiguity in the discussion around what people mean when they say "good code".

Good code for a business is robust code, that's functionally correct, efficient where it needs to be and does not cost too much.

I believe most developers who care about good code are trying to articulate this, they care about a strong system that delivers well, which comes from good architecture.

LLMs actually deliver pretty well on the more trivial code cleanlines stuff, or can be made to pretty trivially with linters, so I don't think devs working with it should be worried about that aspect.

What is changing fast is that last point I mentioned, "that doesn't cost too much" because if you can get 70% of the requirements for 10% of the perceived up front cost, that calculus has changed. But you are not going to be getting the same level of system architecture for that time/cost ratio. That can bite you later, as it does often enough with human coders too.

To me this is semantics as far as it's related to "why don't you want to learn?"

I have a co-worker in another team that write java endpoins we consume. I can tell him what I need and I trust the output. I don't need to know java to trust him, it doesn't mean I don't want to learn.

There are thousand examples like this across every stack and abstraction level. From ssh-handshakes to gps.

Sure my co-worker is fundamentally different from a compiler which is fundamentally different from an LLM.

My argument is that the chain-of-trust where you offload knowledge to an external source is identical. We do it all the time but somehow doing it with an LLM means we no longer want to learn?

Local models are increasingly becoming capable of taking on serious coding tasks that I would have previously sent to a frontier lab

Yes, but that's a completely different argument (that I agree with). Essentially, yes they are conceptually similar but one is bad because you have to pay rent to use it.

Oh managers are not the biggest evil here. At least they know basics.

Now we have influx of people with not a single shred of technical knowledge thinking they can create something.

It ends like this: all codebases become unmaintainable spaghetti after agentic AI spends years on it. Then after every agent in existence will spend minimum 24 hours reading the codebase to add a simple feature, the software is abandoned.