Right now I think there is an edge to how you construct prompts and config files. There is a large difference between "modify f() to do..." and "modify f() to do... Review the current variables and make sure they are still used consistent with their naming. Look for unreachable and dead code. Examine callers and called functions for side effects from the introduced changes...".
I don't think that will make much difference in a year.
I'm increasingly convinced of the opposite. IMO Fable was pretty similarly capable for my day to day work as Opus.
I think there's a pretty good chance that we've reached the point of diminishing returns, for our specific use case.
There are still like a billion other (more difficult) use cases to be tackled, but I think "generating code" has gotten really good to the point where the other bottlenecks will prevent further exponential progress on this specific task.
It's already going away for me in a sense as I build up a library of AGENTS.md and Codex skills. I see no reason such things won't get baked in at the agent layer so that domain specific rules and such are automatically applied when appopriate.
You're essentially making the case here that your work is now automated into a set of one-shot actions that can be performed by an AI model and your job has become to selectively apply these actions. That says either a) we don't do the same work, and instead you're doing some kind of low level devops function that I've only ever seen in rare cases where a human isn't needed anymore, or b) you've vastly oversimplified the software engineering you're doing.
Sophisticated chain of reasoning LLMs like ChatGPT have baked in some natural language operations and they make it so i can create at a higher level of the language expression stack. But I'm still formulating my own expression. There's no conceivable path I can see where an improved model is going to be able to do what I do. I think that is clear from my ChatGPT threads at least.
I think you're reading quite a bit into my comment.. I'll try to respond to a more accurate response to my comment but I'm not going to waste time with this sort of response.
I'm not sure if you're ahead of me or behind me on this curve, but fwiw, my experience has been that we have now encoded everything that is useful in the various markdown files and have reached the point of diminishing returns on this, with more powerful new models making noticeable but not revolutionary improvements as they come out.
Yeah, uh, why would it go away? In what world do you completely surrender your ability to control the work product, the methods for achieving said work product, etc. That is the dream of a PHB.
Not OP, but I generally agree. Models are powerful enough now to reliably instruct other models. They don’t need fancy tools or IDEs, just the command line.
With deterministic workflows, type-safe languages and test suites, agentic loops pretty much “can’t fail”. They will continue until the types resolve, the tests pass, and the project requirements are deterministically met.
By that point it’s literally just a case of typing a prompt in to a text field, and waiting.
If you think that you're simply incorrect. For the average person in a Western country, life is dramatically better anytime in the next hundred years, including the worst possible outcomes of climate change. Not to mention that the exponential growth of solar panels is basically stabilized in climate right now.
> Not to mention that the exponential growth of solar panels is basically stabilized in climate right now.
Sadly, growth of PV can only deal with part of the problem. We're making PV fast enough now that, given panel lifetimes, in 30 years 100% of current electricity demand will be met with specifically PV, and we're also making wind turbines and nuclear reactors and stuff.
The "and also" means we'll probably also be fine electrifying land transport.
Air transport (also fast-and-reliable sea transport) is somewhat harder to make renewable, but theoretically possible. Metal extraction from ore can be done electrolytically.
Concrete's not limited by electricity, it's an independent thing to be solved. So is meat farming. Progress exists on these, yes, but my point is they're not solved just by us being on the home stretch for electricity.
I don't think your "worst possible" case is calibrated correctly.
The worst possible case is global famine and food chain collapse. Food wars, water riots. Starvation in the literal billions.
Living in a Western country won't solve the problems of "crops can't grow anymore" or "keystone species in the food chain are extinct". We'll starve and die in mass numbers just like everyone else.
Even if we had billions dead from starvation worldwide (huge if), life in a western country in 2060 will be better than life for almost anyone born in 1800 anywhere on earth.
The difference is that president Kamacho cared about his people and was ok with being counseled by the smartest guy in the country to turn things around.
I think maybe this is getting hugged to death. I searched for an old favorite of mine: `radix - bright eyes`, and couldn't find it, but maybe I'm just doing it wrong.
Could you say more? I don't really follow, and I've used trackers for a long time. Don't some trackers already have something akin to this in terms of "randomizing" wave forms inside some reasonable parameters? Why would you need AI for this problem?
Maybe I'm just lucky, but I almost never run into paren-matching issues. Hooked to a working REPL, Clojure isn't merely "good", I'd say it is dominant compared to a lot of other langs.
"as most arguments don't apply to today's world" makes me want to roll my eyes so hard at you. The vast majority of problems we had with building complicated systems are all still just sitting there. People are speedrunning relearning things we've known about software engineering for decades.
The more things change, the more they stay the same.
Between AI and the stock market (which of course relates directly to AI), I’ve lost count of the number of times I’ve heard lately another variation of “this time is different.” Sometimes so close to those words that I wonder why the person speaking them doesn’t feel a bit tingly. Great big warning signs all around.
The examples I gave, and the arguments that usually support them don’t really translate into “building complicated systems”. I was talking about the arguments in support of variable naming flamewars, etc.
I’m not proponent of AI generating everything without any supervision as of now. But willing to change my mind when it gets better.
Most software engineering jobs are not cutting-edge tech, or research, or solving unsolved problems. Integrations, APIs, figma-to-react pipelines, devops and etc. is what people get hired for. All those can be done much faster in the same-or-better quality by an experienced person with the supplement of AI. It’s hard to imagine any company would go against the grain and slow things down on purpose.
So I accept that “nonsense arguments are nonsense”, but with some minor differences of opinion. Naming of things matters insofar as you care as a human to actually conceptualize the system you’re building. You can call all of this stuff minutiae, and on some level I kind of agree, except for the general vibe of _caring about the quality of the stuff you produce_. That is something that still matters whether it “works”. Like, yes you can get an LLM to gen some junk, but _is it any good_ is still something you are in charge of.
As far as “boring systems are boring”, I can tell you from experience that I work on a pretty boring system, and AI is not all that meaningful in terms of its impact, and it’s not for a lack of trying.
Can it help me create a migration and add an endpoint and such? Sure. But those aren’t the hard problems. They never were.
It’s funny that you think the idea of slowing down is such a bad one, but it is another well-established truth. Slow is smooth, and smooth is fast. This notion of break/fixing your way to prosperity by way of 10,000 ill-conceived PRs is a fool’s game.
I'm sorry, you might be right. But this simply doesn't reflect my daily reality. All I can say is, nobody in my org is creating 10,000 PRs. But everyone is using Claude Code for virtually all commits. We've been doing it since about Opus 4.5ish. So far, so good.
Generally we've modified our timelines heavily, systems are working as intended, company is still making money. There are some AI-authored commits that had mistakes that we didn't catch, but I'm sure this could've been an issue even if all were human-authored. I know first-hand multiple other companies who are doing exactly the same thing.
I agree with "slow is smooth, and smooth is fast" for mission critical systems. But super majority of systems are, indeed, not mission critical.
I think we're probably talking past one another a little bit. I use LLMs. Daily, even. I've been doing so since around the same time. The vast majority of people in my organization are doing the same.
I have watched some projects absolutely explode in LOC added, number of PRs, etc. but I think the more interesting question is: how much of it is directly being done to add customer value, how much of it is churn, etc. you might get some interesting answers.
As so frequently seems to be the case for you and I, we kind of agree but then you drop something that just does not compute for me: "slow is smooth, and smooth is fast" is not specific to "mission critical" systems, it is generally applicable.
As I said in a previous comment, I work on a fairly boring system. Its "criticality" is debatable, but in general we make the same kinds of boring guarantees to our users that even mediocre SaaS products offer: a few 9s of uptime, zero-downtime deploys, etc. AI has made aspects of working on this system easier, but in terms of API surface, how users are using it, how to safely advance its state without breaking existing callers, data migrations across services, and so on, very little has changed.
It depends :). It’s enough I pay for it for my silly side project. Historically we’ve paid a lot for software tools. IDEs and even documentation used to be pretty expensive. AI seems at least on par with those.
reply