My views on AI use

Mindful of the ways in which it is bad, I am choosing to write this completely by hand. My thoughts on AI are many and varied. Let’s start with the bad.

¶The bad

This is a list of bad things on the top of mind. It’s not exhaustive and definitely not ordered by priority or seriousness. It’s just what I think of when pondering why not to use AI. These things kept me from not using AI for the longest time.

¶AI is built on stolen work

This is the thing I identify with the most. AI is built on stolen work. My work. Your work (assuming you’ve written any code or books or social media posts). People whose work I like, and people whose work I don’t.

It’s a heavily litigated fact that these large AI labs have bought and stolen access to books, scanned them into the large plagiarism machines, and often times destroyed the books afterward. I doubt many people have read or even know of the many books I have written. I doubt the labs have stolen much revenue from me.

The point is they never asked.

The same can be said for my code. While most of my public code has a permissive MIT license, the labs creations were built by ignoring licenses. Doing exactly what the licenses prohibit (dropping attribution, removing license files, copy-left etc).

The large models we use every day would not exist without this theft.

¶AI threatens jobs

It’s not enough that the labs have stolen the code we’ve written in our meagre “off time”. They are actively selling the idea that our employers can replace us with cheaper robots that don’t come with pesky strings like sleep and rights.

They are obviously over-selling the robot replacements, though. At least in their current form: LLMs are incapable of replacing me and the state of the art harnesses will only make their job harder.

Where are the robot dishwashers and bricklayers? Why are all the fulfilling jobs being threatened?

¶AI is exacerbating negative climate change

Duh. More power spent generating memes and sloppy code means more CO2 in the atmosphere.

No amount of greenwashing fixes that. The only path forward I can see is nuclear, fission, or fusion energy. I hope AI forces developments, here, because (at least) nuclear energy is a great way for us to leave fossil fuels behind.

Right now, the desire for tokens is pushing the climate in the wrong direction. The desire for more ram is pushing the climate in the wrong direction. The need for more GPUs is pushing the climate in the wrong direction.

¶AI is not that good (yet)

I hinted at this, but it needs its own section. I use AI all the time. My informed opinion is that AI is not good enough to replace a skilled worker. A skilled worker can achieve a much better quality of life using it effectively; but firing the skilled worker is a mistake.

It will take decades for greedy corpos to understand this, and in those decades I will make a killing replacing all the slop low-skilled and/or careless people have pushed to prod. The talking heads on twitter would have you believe we can abandon all the lessons 50 years of software development has taught us; but they are trying to sell you the cure to the skill attrition they are trying to bring about.

Just this week I read a thread where someone, selling a tool that containerises every feature branch/work-tree, is evangelising “not using branches anymore”. Ok, buddy. You sell your thing.

¶AI is not sustainable

The current landscape is changing really fast.

Anthropic tokens cost 3x-8x less through subscription than they do through the API. Remember when Uber rides used to be cheap? Then, they killed enough competition to be able to jack up the prices. That is what capitalism does to things without competition.

What happens when we’ve let these tools make us comfortable and dull? When we’ve given over our entire workload (assuming we’ve not been replaced with AI), and a Claude Code outage means we can’t do any work?

You know and I know that we’re not supposed to hand over our thinking to an agent, but the labs are pushing the narrative that the agents can replace people. They’re telling our boss we should have all the responsibility of what the AI produces but none of the agency in how and when to use it.

Ok, that is far enough into tinfoil hat territory.

¶AI makes me tired

It does this in many ways, but the big causes are:

1. It makes many of the same mistakes over and over. It is not good at adapting. It is just as capable as it was yesterday, and it will not improve without constantly upgrading to a newer model or constantly upgrading the tools and skills it uses.
2. The natural regulator that was “this task is too big to start at 8pm, I’ll leave it for tomorrow” is no longer in play. Highly motivated people are working more because they can take on bigger and bigger things. Work-life balance is more important; but we ignore it because side-projects are on the life side, right?

There are some small gains to be made in both of these areas, but they were much smaller issues before AI. I didn’t think I would have to re-learn how to solve them this far into my career.

¶The good

I use AI every day. It drafts 100% of my code. Without it I would have a lower quality of life. I am privileged enough not to have to think too hard about that.

¶AI is good at drafting code

This seems like it runs contrary to the “AI is not that good (yet)” section; but hear me out. With AI, I am able to make more and better things than I could without AI. I have 20 years of experience making things without AI and, while I am more tired now, I am better with it in my life than without.

Before AI, the mythical 10x developer’s strength was shipping fast. Now, everyone can ship the biggest pile of crap imaginable. I don’t think it’s actually possible to produce 10x as much work if you care about quality.

Instead, I have worked hard to build workflows and skills that allow me to produce closer to 2x-3x the amount of quality code I was making before, and using the rest of the slack to enjoy my life. Regardless of when/how the bubble bursts; this is the only sustainable path forward for me.

¶AI is good enough at translating

I’m not just talking about translating text to other languages, though it is good enough at that too. I’m talking about translating patterns and architecture to new environments. I’m talking about translating how things new things work in a way that appeals to my situation and experience.

It’s not as good as skilled workers are at this kind of translation; but it is good enough to make many things that were gated by time and desire open to significantly more people. A Spanish kid just getting into programming can read the tutorials for my toy game engine without first learning English. I can build apps that make life easier for my family without first learning Swift and Kotlin. These are good for me. I can only imagine how much easier the world just became for people not similarly temporarily abled or similarly privileged.

¶AI is pushing caring people toward doing more good

Despite all the worst things corpos are using it for, caring people are being pushed to doing more good in the world. More people are making space to be unplugged, or growing physical media collections, or showing how they make amazing things without the help of AI.

They shouldn’t have been put in this position to begin with; but it’s amazing to see creativity blossoming in the world. These people aren’t rejecting the latest model or meme generator. They’re rejecting the idea that people should be lazy and unlearned. They’re embracing the difficult processes of learning being vulnerable in their imperfect state.

My daughter’s dislike of all things AI has pushed her to make the most beautiful things. My wife’s dislike of slopvertising has pushed her to do more pro-bono design for community businesses. My dislike for website UI generator apps has pushed me to learn more about designing simple interfaces.

¶This isn’t settled

I can see myself adding to these lists and refining things as the landscape changes. AI is not all bad or all good. I have learned it’s safe to ignore opinions built on these axioms. I have struggled on both sides of this debate, and my views grow more nuanced as time goes on. I think that’s healthy.

I have written and deleted many thoughts, and nowhere more than this conclusion section. I hope the current labs are wiped out in lawsuits and bubble bursting. I hope the labs that replace them are driven by human good instead of capitalist nihilism.