We’re burning the planet to generate code I have to reject

I have been debating about AI with a lot of people recently, and I wanted to make one big post to get rid of my thoughts and hopefully spark a discussion.

I come from a programming background. From what I see, the people who were hand writing bad code still are doing the same but with LLMs, now it’s just more of that, faster. I think that’s a huge part people don’t like. As soon as you give an LLM too much freedom and don’t guide it through every single choice, it writes horrible code. It can speed things up if you already designed what you are going to make, and know how to write it.

In my own experience, I have to review a lot of code at my job, and of the submissions that get turned down, about 90% are rejected because they’re AI slop. I much prefer hand-written code with a few bugs in it that I can just fix. Or people properly thinking about something before telling AI to just do it.

In another discussion I had, this got mentioned, and I think it’s very suitable here:

“Early on in the LLM race, I spoke with a guy who worked at one of the now big AI companies. He echoed essentially what you said, that the LLM is a multiplier. ‘If you write good code, you get more good code. If you write shit code, well you get a pile of it.’”

I completely agree that some people are able to multiply their productivity by crazy amounts without losing attention to quality or their skill sets. However, I think this is a very small fraction compared to the overall output we are seeing at companies and on the internet. A lot of people are outputting things in days that would’ve taken them weeks, but the quality of a lot of that work goes down.

I think a big trend we are now seeing is people trying to reach further than they previously could, or had the knowledge/skill set for. Because LLMs are like an interactive dictionary, we are less limited in what we can accomplish. But in doing so, we often leave holes in our approaches without realising. If we have to learn/research everything from scratch, we are forced to look at every part that comes with it. With AI, we now have the freedom to look over some of these important details, which seems very fragile to me.

I also think LLMs have a possible negative impact on our education system. Educational institutions will have a lot of trouble integrating this new way of working, and during the initial years it will only make it harder for students to actually learn new things and be prepared for their future. Both because it’s now very unclear what the future will look like, and because people can easily become lazy and not learn important fine details.

Then there’s the hype. It’s hella cool that we’ve got all of this; 10 years ago this would’ve been unimaginable. But people treat AI as something it isn’t yet, and, given current algorithms and technology, likely won’t be. It’s still just a very primitive box, and to quote someone I discussed this with, the hype “paints a false narrative of where we are technologically.”

That hype drives large companies to pour enormous resources into a technology I don’t think they fully understand. 95% of actually useful LLM automation or daily usage would be fine with a model that came out last year or earlier. Yet each big company is rapidly trying to best the others with a model that is “slightly” better, requiring 10x more resources, and already hitting a big plateau. If it were not such a capitalist race for market share, we could’ve worked together and gotten much further, much quicker, with way less money and resources. Of course, this is an unrealistic utopia.

One could argue that AI boosts our productivity, and that this in turn benefits our lives and the world. I hope the gains outweigh the costs, but right now I have my doubts. Especially seeing the state of our planet, I’d much rather see that money and talent spent on directly improving the planet and society. But unfortunately, that doesn’t have monetary gain for the huge companies who possess most of the wealth.

Long story short: I think AI can be great, but the slop is out of control and the downsides are real. People need to see LLMs for what they are, not a magic fix that spares us from doing the work. And meanwhile, we’re burning through a staggering amount of resources to make it happen, especially when it comes to our planet. That’s energy, water, and money we could be spending on things that matter a lot more.

Edit:

I am not anti-AI. I really believe this is awesome, and some people, with the right mindset, scaffolding, way of working, harness, and knowledge, are able to accomplish great qualitative and productive workflows! I just think that, looking at the whole picture, it has a lot of negative downsides too. I think those downsides deserve attention as well. Also, while LLMs are great, I think people are overestimating them and treating them as more than they actually are.

Secondly, I don’t think AI is the biggest environmental problem we face. I just see so much money and so many resources being wasted. It would be so awesome if humanity could put that same amount of money, resources, and hype into directly helping the world and the environment! It’s crazy to see what we can accomplish as a collective, just imagine that same collective effort directed toward a concrete goal like solving hunger, environmental issues, etc.

AI itself isn’t the environmental problem, it’s the sheer scale of money, resources, and hype being funneled into it that feels wasteful, especially when that same effort could be directed at hunger, climate change, and other urgent needs. Spending this much on anything that doesn’t directly benefit the world is hard to justify.

The title’s a bit provocative, but hey it worked:)

Edit 2:

Environmental concerns aren’t the only argument people have against AI. Consider:

Everywhere you look now, there’s poor-quality AI content, whether it’s something a company has put out or a post on social media. People are pumping out low-quality material incredibly fast because of AI.
Poorly written software.
Plagiarism.
Leaning on AI for thinking, writing, or problem-solving can weaken people’s own skills and critical thinking. And in school this can hurt learning itself, students who outsource essays or homework may pass the assignment without actually absorbing the material, leaving gaps that show up later in exams, higher-level courses, or the workplace.
AI makes scams, phishing, cyberattacks, and harassment possible at scale.
Misinformation and deepfakes: AI makes it cheap to produce fake images, voices, videos, and convincing false text, which erodes trust in what’s real.
Confidently stated errors: AI can “hallucinate,” giving wrong facts, citations, or advice that sound authoritative.

And even on the environmental side, you have to factor in chip and hardware production, not just the impact of running the servers. There are plenty of serious arguments against AI.

My biggest personal frustration is the quality gap between AI-generated work and the work of a skilled person who has invested real time and effort. I see it daily at the software company where I work: without human oversight of every small decision, AI output is consistently poor, and the result is a lot of low-quality work being produced everywhere.

Of course it can speed some things up, and occasionally it handles a detail better than you would have. I’m not saying it isn’t a great tool. But the overall quality still falls short.

People make it out to be more than it is: it’s not a superhuman, it’s a tool. The hype paints a false picture of where we actually are technologically. And some people are rightly against it, because it comes with real downsides too.

And just because AI isn’t the leading cause of environmental damage doesn’t mean it’s harmless either.

Checking code takes so much time now, I once read this comment that rings very true:

“The review cost isn’t just ‘more code to read’. AI-generated changes often look idiomatically clean while quietly getting ownership or lifetime semantics wrong in ways that only surface under specific conditions. Skimming used to work because a human author’s mental model was baked into the structure; now you have to actively reconstruct that model, because it might not exist at all.”

- Joshua