Why I remain a skeptic

Posts

Tags

RSS Feed

Codeberg

GitHub

Mastodon

YouTube

E-Mail

Why I remain a skeptic

14 Aug 2026

| ai

I still do not use LLMs for anything I actually care about. There are a lot of reasons I<br>could give: environmental, social, political. I could talk about ecosystem lock-in,<br>learned dependence, self-soverignty, philosophical concerns, the intentional dismantling<br>of labour bargaining power, and more. But before all of that, the simple question of<br>efficacy - whether they actually work for the purpose of building non-trivial software -<br>is still very far from settled.

Here are a few reasons for my skepticism:

We are 4 years into this ‘revolution’ and the industry has almost nothing to show<br>for it. Software has not improved in quality, got faster, become cheaper to produce<br>(when you exclude the mountain of poor-quality demoware that no reputable organisation<br>would touch with a barge pole), or become more capable. It’s certainly not become more<br>secure. A technological revolution is usually marked by old empires falling and<br>being replaced with new players and ideas. It is not clear to me that the industry has<br>produced any new ideas in the last few years that haven’t been in the navel-gazing<br>arena of AI boosterism. Everybody is yelling at the top of their lungs but nobody is<br>actually shipping anything useful.

I do not feel like I’m falling behind by not using them, despite expecting to.<br>Nobody has turned up to steal the reputational lunch money of my open-source projects,<br>for all the noise. You might be able to replicate the function of my code, but so can<br>a simple memcpy: an open-source project that people feel they can trust enough to<br>build their system on is about so much more than a feature list.

The evidence is tenuous and thin. $1.5 trillion later and we still have almost no<br>independent studies that can attest to top-level productivity improvements associated<br>with AI. Given that half of our economy currently rests on this central claim being<br>true, we should expect more than a few one-off studies and anecdotes. Every study I’m<br>aware of either zooms in too close and tracks a metric that has nothing to do with<br>top-level productivity (‘lines written’, ‘PRs merged’, ‘features added’, etc.) or is<br>too small to draw useful conclusions from. Even then, those studies that do exist and<br>suggest some amount of academic righor point to only marginal or even negative<br>productivity gains.

The LLM-generated PRs I get are still rubbish. The code has become less obviously<br>repetitive and less stylistically consistent, but no more fit for purpose. The<br>proportion of them that are worth actually merging is still much lower than those<br>written by humans. A fair criticism is that it’s becoming increasingly difficult to<br>tell the two apart due to AI outputs becoming slightly less formulaic, but even the<br>most careful humans will still leave the odd typo in their description so I’m<br>unconvinced that the bots are yet walking among us to that degree.

They still miss obvious shit. Several of my open-source libraries are in the<br>dependency trees of FAANG companies and I’ve had several reports passed to me as a<br>result of scans with frontier models. Some of it’s good, most of it’s bad; I can<br>recall several situations in which a frontier model missed something obvious and<br>important that was found by a hobbyist using their brain soon afterward.

The philosophy exuded by AI boosters still defies the last half-century of theory<br>about what software development even is. We should all know by now that code is an<br>input to the software development process and not an output: yet even the most<br>forward-looking organisations still tactitly treat lines of code as a productivity<br>metric. That is the path to unmaintainable slop, whether the slop comes out of a<br>machine or is simply the byproduct of humans conforming to poorly designed incentive<br>structures. Peter Naur remains<br>undefeated.

When everybody is super, nobody is. The not-so-quiet goal of the LLM push by big tech<br>was to homogenise intellectual labour such that it can be made fungible. When labour<br>is fungible you can swap it out whenever you like for an alternative, decreasing the<br>bargaining power of workers. Software development was one of the last ‘guild<br>fields’ and now we are seeing its industrialisation occur. By not leaning into<br>LLMs I’ve given myself the room to specialise in a smaller set of skills. So far that<br>conscious decision is working out well, and I remain confident about my job security.

Enter the tarpit 🤖

If you notice accessibility issues with this site, please let me know!

© Joshua Barretto