Believe the Checkbook

Each of the last productivity multipliers coincided with greatly expanded markets (e.g. PC revolution, internet, mobile). Those are at the saturation point. And we've effectively built all the software those things need now. Of course there is still room for innovation in software, but it is not like in the past where we also had to build all the low hanging fruit at the same time. That doesn't require nearly as many people — and that was already starting to become apparent before anyone knew what an LLM is.

This AI craze swooped in at the right time to help hold up the industry and is the only thing keeping it together right now. We're quickly trying to build all the low-hanging fruit for it, keeping many developers busy (although not like it used to be) but it currently doesn't look like there is much low-hanging fruit to build. LLMs don't appear to have the breadth of need like previous computing revolutions had. Once we've added chat interfaces to everything, which is far from being a Herculean task, all the low-hanging fruit will be gone. That's quite unlike previous revolutions where we had to build all the software from scratch, effectively, not just slap some lipstick on existing software.

If we want to relive the past, we need a new hardware paradigm that needs everything rewritten for it again. Not impossible, but all the low-hanging hardware directions have also been checked off at this point.

Demand for software has been tied to demand for software engineering labor. That is no longer true. So demand for software may still go up while demand for labor goes another direction.

Also I do hold a belief that most tech companies are taking a cost/labor reduction strategy for a reason, and I think that’s because we’re closing a period of innovation. Keeping the lights on, or protecting their moats, requires less labor.

No shade on your skills, but for most problems, this is already false; the solutions have already been scraped.

All OSS has been ingested, and all the discussion in forum like this about it, and the personal blog posts and newsletters about it; and the bug tracking; and theh pull requests, and...

and training etc. is only going to get better and filtering out what is "best."

"Writing code" is not the goal. The goal is to design a coherent logical system that achieves some goal. So the practice of programming is in thinking hard about what goal I want to achieve, then thinking about the sort of logical system that I could design that would allow me to verifiably achieve that goal, then actually banging out the code that implements the abstract logical system that I have in my head, then iterating to refine both the abstract system and its implementation. And as a result of being the one who produced the code, I have certainty that the code implements the system I have in mind, and that the system it represents is for for the purpose of achieving the original goals.

So reducing the part where I go from abstract system to concrete implementation only saves me time spent typing, while at the same time decoupling me from understanding whether the code actually implements the system I have in mind. To recover that coupling, I need to read the code and understand what it does, which is often slower than just typing it myself.

And to even express the system to the code generator in the first place still requires me to mentally bridge the gap between the goal and the system that will achieve that goal, so it doesn't save me any time there.

The exceptions are things where I literally don't care whether the outputs are actually correct, or they're things that I can rely on external tools to verify (e.g. generating conformance tests), or they're tiny boilerplate autocomplete snippets that aren't trying to do anything subtle or interesting.

In my experience (and especially at my current job) bottlenecks are more often organizational than technical. I spend a lot of time waiting for others to make decisions before I can actually proceed with any work.

My judgement is built in to the time it takes me to code. I think I would be spending the same amount of time doing that while reviewing the AI code to make sure it isn't doing something silly (even if it does technically work.)

A friend of mine recently switched jobs from Amazon to a small AI startup where he uses AI heavily to write code. He says it's improved his productivity 5x, but I don't really think that's the AI. I think it's (mostly) the lack of bureaucracy in his small 2 or 3 person company.

I'm very dubious about claims that AI can improve productivity so much because that just hasn't been my experience. Maybe I'm just bad at using it.

All you did was changing the programming language from (say) Python to English. One is designed to be a programming language, with few ambiguities etc. The other is, well, English.

Speed of typing code is not all that different than the speed of typing English, even accounting for the volume expansion of English -> <favorite programming language>. And then, of course, there is the new extra cost of then reading and understanding whatever code the AI wrote.

I think OP is closer to the latter. How I typically have been using Copilot is as a faster autocomplete that I read and tweak before moving on. Too many years of struggling to describe a task to Siri left me deciding “I’ll just show it what I want” rather than tell.

> Or are you saying that you believe you can't get that code written without spending an equivalent amount of time describing your judgments?

It’s sort of the opposite: You don’t get to the proper judgement without playing through the possibilities in your mind, part of which is accomplished by spending time coding.

Klarna, now there's a company that seems to have no idea what direction it's going in. In the past month, they've announced they're going to be at the forefront of Agentic AI for merchants so... agents can figure out what merchants are selling? They're somehow offering stablecoins to institutional investors to use USDC to extend loans to Klarna? And then they're starting some kind of credit-card rewards program with access to airline lounges?

As someone not close to retirement yet, it's a very sad shift.

Anything you do with AI is improved if you're able to traverse the stack. There's no situation where knowing how to code won't put you above peers who don't.

It's like how every job requires math if you make it far enough.

> I explain to them that typing speed is never the bottleneck in the work that I do.

Never is a very strong word. I'm not a terribly fast typist but I intentionally trained to be faster because at times I wanted to whip out some stuff and the thought of typing it all out just annoyed me since it took too long. I think typing speed matters and saying it doesn't is a lie. At the very least if you have a faster baseline then typing stuff is more relaxing instead of just a chore.

These are all things I'd rather have seen the article set out to talk about as well, instead it opens up to disprove a statement that AI can write the code by means of showing it being used that way with Bun.

> Where will these engineers get their experience in the future

The house of the feeble minded: https://www.abelard.org/asimov.php

[delayed]

I was thinking the same but it's like they only used AI to handle the editing or something because even throwing it into ChatGPT "how could this article be improved: ${article}" gives:

> Tighten the causal claim: “AI writes code → therefore judgment is scarce”

As one of the first suggestions, so it's not something inherent to whether the article used AI in some way. Regardless, I care less what did the writing and more whether the result was good.

This seems overly pedantic. The intended meaning is clear.

Deep learning and transformers have given step functions in AI's capabilities. It may not happen, but it's reasonable to expect another step-function development soon.

When you have that hair raising “am I crazy why are people touting ai” feeling, it’s good to look at their profile. Oftentimes they’re caught up in some ai play. Also it’s good to remember yc has heavy investment in gen ai so this site is heavily biased

Unfortunately this is a skill issue. And it's a totally different skill than reading and writing code, building solid systems, and general software engineering at large. That is annoying, but where we're currently at.

Assume you're writing code manually, and you personally make a mistake. It's often worthwhile to create a mechanism that prevents that class of mistake from cropping up again. Adding better LSP or refactoring support to your editor, better syntax highlighting, better type checking, etc.

That same exact game of whack a mole now has to be done for you and whatever agent you're building with. Some questions to ask: What caused the hallucination? Did you have the agent lay out its plan before it writes any code? Ask you questions and iterate on a spec before implementation? Have you given it all of the necessary tools, test harnesses, and context it needs to complete a request that you've made to it? How do you automate this so that it's impossible for these pieces to be missing for the next request? Are you using the right model for the task at hand?

I've noticed some of this in recent months. I've also noticed people editing out some of the popular tells, like replacing em-dashes with commas, or at least I think so, because of odd formatting/errors in places where it sounds like the LLM would have used a dash.

But at this point I'm not confident that I'm not failing to identify a lot of LLM-generated text and not making false positives.

>instead there is a societal shift to better recognize it

Unlikely. AI keeps improving, and we are already at the point where real people are accused of being AI.