Post #2620986 · Elektrine

2026-01-05 10:13 UTC

@HalvarFlake@mastodon.social I feel like the scale of very large projects in software (which I assume you're talking about with "ambition") isn't limited by the ability to write code, but the ability to understand the whole system as a group, communicate it within the group, and set up good incentives within and outside the project. None of these are helped with 2025-era LLMs, and I imagine not 2026's either. Building rickety structures faster seems unhelpful. Is my logic flawed?

Replies (5)

@karl@infosec.exchange 2026-01-05 10:28

@headmold@mastodon.social @HalvarFlake@mastodon.social no, it's not. But I've seen many instances pre-LLM of people in charge with just the right mix of insight and cluelessness to state that doing change X / feature Y is just a matter of a few rickety structures here and there despite the software engineering team making it clear that it isn't the way to go. My point is, I don't trust people in charge to understand why building rickety structures is unhelpful, so to them LLMs are very much looking just like a huge boost in productivity.

Open ##2620988
@markhughes@mastodon.social 2026-01-05 13:51

@headmold@mastodon.social I find it odd that with so little to show for the past two years, and lots of evidence to the contrary (which makes sense to me), people are still making incredible claims for #LLMs. I've been at the cutting edge of software development for decades, alongside AI projects in the 1980s (and managed one). I've followed AI with optimistic interest, dabbled, thought hard etc. Even considered setting up an AI lab. I've also seen snake oil come and go. This stinks of it. @HalvarFlake@mastodon.social

Open ##2620989
@markhughes@mastodon.social 2026-01-05 13:51

@headmold@mastodon.social I find it odd that with so little to show for the past two years, and lots of evidence to the contrary (which makes sense to me), people are still making incredible claims for #LLMs. I've been at the cutting edge of software development for decades, alongside AI projects in the 1980s (and managed one). I've followed AI with optimistic interest, dabbled, thought hard etc. Even considered setting up an AI lab. I've also seen snake oil come and go. This stinks of it. @HalvarFlake@mastodon.social

Open ##2620990
@markhughes@mastodon.social 2026-01-05 13:54

@headmold@mastodon.social I find it odd that with so little to show for the past two years, and lots of evidence to the contrary (which makes sense to me), people are still making incredible claims for #LLMs. I've been at the cutting edge of software development for decades, alongside AI projects in the 1980s (and managed one). I've followed AI with optimistic interest, dabbled, thought hard etc. Even considered setting up an AI lab. I've also seen snake oil come and go. This stinks of it. @HalvarFlake@mastodon.social

Open ##2620991
@trademark@fosstodon.org 2026-01-05 14:55

@headmold@mastodon.social @HalvarFlake@mastodon.social I think it is, yes. While I consider the current LLMs not good enough for production quality code. They are more than good enough for various auxiliary tasks: experiments, debugging helpers, test harnesses, status pages, custom dataset explorers, impedance matching, fixing the build system. This vastly increases the scale of the projects you can take on. The actual production code is a small part of any large project.

Open ##2620992