@dave

The industry never got to the point of figuring out how to solve the same amount of problems with less work.

Now the opposite is happening.

The ratio of software developers to available work was already a bottleneck.

If you consider it as a problem of energy efficiency, this was inevitable. As far as I can tell, there is no increase in total energy efficiency for the output of quality work.

And that's what makes it different from other technological revolutions.

@dave she views AI merely as a "speed multiplier" for chores, missing that it is actually a capability unlocker. She ignores that software complexity has outpaced human cognitive limits; without AI, managing or refactoring today's massive codebases is becoming impossible.
We aren't just washing dishes faster; we are building systems—like complex biological modelling—that a human team simply couldn't execute manually. The "stress" is the friction of rapid innovation, not a bubble.

@dave I keep seeing takes like this…and I’m wondering if I’m really bad at coding and naive, or people are in denial of just how effective LLM is at it.

I barely have to write syntax directly these days. I often have to prompt for improvements, and use some of my own knowledge with debugging, but even after all of that, I feel like my productivity is many times faster since December.

I lean towards the skepticism of this tech being more bullshit than the claims that it’s revolutionary.

@dave maybe part of why LLM is working so well for me is I’m focused on a green field code base only @suorcd and I maintain right now. With a large convoluted legacy code base that has changed many hands, I imagine LLM may not work so well, but it seems to cut right to what we need done.

That said, we’re only in Alpha with new Podverse and haven’t been truly tested with traffic yet. It seems like a far more robust and clean setup tho and I’m expecting / hoping it will be a major improvement.

@mitch @dave @suorcd

You're also working on really well-known problems in a very popular programming language.

@dave I hit the burnout phase this week. LLMs are an awesome tool for coding but you have to learn to focus your energy on 1 thing.

@dave@podcastindex.social Such a great video to watch. AI is a hot topic right now and this should be watched.

@mitch
The trick is code has strict syntax rules, unlike prose. Remember that people aren't just using LLMs to write software, and that's going to color their experiences using LLMs
@dave

@matt_the_tall @dave true. I saw "The Tech Report" and assumed it was mostly a SWE audience.

@mitch
I've seen examples of people using LLMs to create BS code, too, so source code isn't immune to the issue
@dave

@mitch
Probably worth mentioning the channel has only 20k subscribers and it only showed up last year, so I wouldn't make assumptions about their audience
@dave

@matt_the_tall @dave it certainly can produce garbage. It just seems like it can be steered effectively for dramatic productivity boosts in programming.

I'm throwing a pretty complex plan at Codex 5.2 today for v4v integration + metaboost + boostbox + alby prod/sandbox implementation. will see how it goes...