Concerns about the actual state of LLM tools

In this article I'll try to pin down what irks me about LLMs (Large Language Models, sometimes called "AI") tools, a topic that seems unavoidable if you are in IT. I want to take a screenshot of what's in my head at the time of writing (February 2026) in order to compare what will happen in the next 12 months. In the past 12 months there was a tangible evolution in the output of these tools but I am still not impressed for a number of reasons. I am open to be proven wrong.

I also write down this because I am tired of giving the same answers over and over (so if you received this link from me, please don't take it personally).

§ Premise

I don't use any LLM assistance. I regularly (but not frequently) use https://deepl.com as support for small translations. I don't know if that qualifies as an "LLM", I didn't investigate the underlying technology. If that makes you - dear reader - think that I am not entitled on an opinion on LLM tools, feel free to close the browser tab.

§ What irks me about LLMs

These are technical and non-technical arguments that to this day are preventing me to endorse or use these tools:

• LLMs are not a democratic piece of technology: the cost of developing these tools is only for deep pockets, with the result of creating monopolies. "FOSS" LLM engines are subpar and the gap is only increasing.
• I argue that writing FOSS software using such proprietary tools is not really writing FOSS. This is not the same as using Sublime Text instead of Vim (there are unresolved questions about copyright)
• I argue that it's dangerous to endorse a proprietary cloud service that works on a per-request basis. This creates an undesirable dependency in your work. I mention an interesting comment when Claude went down the other day (GitHub issue tracker).
• Besides the fact that people are obviously introducing a liability in their business, using thee tools is putting at a disadvantage those who cannot (or cannot afford) paying the subscription. I am thinking for example to nations under USA sanctions. Students needed to pirate Adobe Photoshop to learn a tool that the market required and then, maybe, paid for a license when that could be covered by an income. LLMs cannot be "pirated" and free models (I am told) are inferior. Local hosting an LLM still requires beefy hardware.
• Disrupting the hardware market: dubious market practices from OpenAI¹
• The annoying rhetoric of companies and enthusiasts alike about this technology is frankly off-putting ("It's here to stay", "learn it or be left behind")
• The way this technology works basically boils down to: get as much data as you possibly can and throw them at a warehouse full of GPUs². This means that whoever has more data and more GPUs wins the game. This triggered an unscrupulous and unregulated race to scraping content without any consideration of bandwidth costs for the hosting servers

§ Second-order effects

"Second-order effects" is a term sometimes used by the cool kids on Hacker News, it's borrowed from MBA books and is a way to say "uh-oh we didn't think about that". Instead of trying to fix these issues they indirectly cause, AI companies are like "oh well, someone else will".

• Lowering the bar for writing code enables a wave of incoming patches to FOSS projects. Either from unskilled and lazy contributors (thinking they can vibecode their way into successful contributions) to classic spam from malicious actors. This is all additional burden on projects that are not equipped to handle this scale of interactions. cURL stopped the bug bounty program, notable FOSS projects needed to pen anti-AI policies (among the many I'll cite Servo, LLVM, nexttest-rs).
• A number of companies, state entities and the general public are using these tools for utterly deranged purposes (Palantir, American ICE, political parties, etc.)

Many argue that technology is neutral and can be used for good and for bad. While technically correct, this is completely unhelpful because leaves a big hole about what and who should do something about it when shit happens. "Technology is neutral" ends up being a free pass to do anything.

§ Non-technical arguments

This is another category of possibly unforeseen problems that are intrinsically compounded in this technology today and it's not clear if there will be a solution. I split these from the technical ones because, as a developer, people in my circles tend to shrug them off; for a number of reasons, we don't want to adventure in fields we're not familiar with. Others acknowledge these problems but since they don't have a solution ... they just keep on ignoring the problem and wave it off as "temporary", trusting some unspecified market or technological force to eventually fix it. Pure genius.

But we are human beings and we are called to think about complexity and how topics are entangled. We don't need to become experts on everything before being entitled to an opinion but we should ask ourselves questions. They're there, they're part of the whole and we can't ignore them.

• Environmental sustainability of this technology. Training these LLMs is currently extremely expensive in terms of hardware and electricity. There is an arms race to get hold of big sources of electricity at the expense of the environment. In this video from 2024 Eric Schmidt (now retired ex-CEO of Google) says:

  And I, [..] in the spirit of full disclosure, went to the White House on Friday and
  told them that we need to become best friends with Canada because Canada has really
  nice people, helped invent AI, and lots of hydropower. Because we as a country do
  not have enough power to do this. The alternative is to have the Arabs fund it. And
  I like the Arabs personally. I spent lots of time there, right?

• Exploiting underpaid workforce for cleaning datasets to feed the training

• Unresolved copyright issues that nobody knows how to regulate

• As a European citizen, I am worried about relying on this technology siphoning huge amount of data. We are (slowly but finally) realizing that.

• People led to think they will lose their jobs to AI. This induced panic does not have any grounding, just a lot of speculations so here's mine as well from reading the book "Bullshit jobs" (2018) by David Graeben (I wrote about that, see my blog post). In short, the thesis is that there are indeed people doing useless (or disposable) jobs but because of how our capitalistic society works, we cannot afford to keep these people at home doing nothing. People without income don't buy groceries, cars, don't go to cinemas or book holidays (in addition to other catastrophic consequences on mental health). On a big enough scale, this would block economy.

§ Arguments in favor

So, I want to enumerate the arguments I hear that are in favor. Enthusiasts using these tools basically say two things:

• My productivity is at an unprecedented level, I can't go back to writing code as before (the "future is here" argument)
• AI is here to stay, whether you like it or not (the "inevitability" argument)

The first argument is somehow still subjective but I concede to be true: some people I know and trust are objectively experiencing this shift (though I have the feeling that many others are not "measuring" correctly). The second one mirrors what these companies selling that product want us to think. Sort of the "have fun staying poor" from the cryptocurrency speculation bubble of a few years ago.

In short: it all seem to be about personal advantage and scifi-esque scenarios of knowledge acceleration, all the rest be damned. Am I missing anything?

§ Conclusions

Even excluding the non-technical issues (something that I find inherently wrong), I am not yet sold on this technology and would rather wait at least until this speculative bubble will finally burst.

In addition, discussing a topic without taking into account the environment where it is born and its effects is not very wise. If we want to understand the future direction that something could take, we need to analyze how that thing came to be, who has interests in its success and then do some math without letting the bullshit add noise to the discussion.

All I can do for now is trying to filter out this topic and avoid it taking mental space in my already overstimulated brain.

See you again at the end of the year!

--