AI can only solve Boring Problems
By David Buchanan, 16th December 2025
And maybe that's fine?
I thought I'd write down my opinions about AI-assisted programming. Not because it's the trendy thing to have opinions on (ok maybe a little because of that), but because I expect to be wrong. I want to look back on this article in 10 years time and laugh at how wrong I was.
https://commons.wikimedia.org/wiki/File:Desertmirage.jpg
It's easy to say that all the hype is bullshit, but as with most things the truth is probably more nuanced. We won't be able to reliably distinguish the hot air from the genuine innovations until after the hype has cooled down. Mountains of capital can prop up lies for a long time (and there are mountains of capital involved).
This article mainly describes the status quo as I see it, but I do indulge in some futurology at the end. It is handwavy throughout, so don't expect any rigorous analysis here.
Boring Problems
As an extension of the idea of "Boring Technology", I propose that there are Boring Problems and Interesting Problems.
Boring Problems are those for which a Boring Technology solution already exists. All other problems are Interesting.
Perhaps you can solve an Interesting Problem through composition or orchestration of Boring Technologies, but at that point the solution as a whole becomes Interesting.
Most problems we encounter are Boring (or will become Boring), but the frontier of innovation in any given field will always be Interesting.
And I don't mean to imply that Interesting problems are "better". If all your problems are boring, you're winning.
Thesis Statement
Returning to the title of this article: AI can only solve Boring Problems.
Given what I just said about most problems being Boring, this is a good thing for AI, right?
AI can handle the "boring stuff" while we humans get to do the "interesting stuff"?
Sounds great???
Maybe, but I have several reservations.
Solutions to Boring Problems are not necessarily Boring
I said that AI can solve Boring problems, but that does not necessarily mean it'll land on a Boring solution. There's a good chance it'll come up with something overcomplicated.
Maybe it doesn't matter, so long as it works. But when it comes time to extend that solution to deal with new requirements, iterating on an Interesting codebase is an Interesting Problem. Oops!
Humans have a finite capacity for Interesting work
The bottleneck is still figuring out the Interesting parts. Having less Boring work to do doesn't necessarily help overall productivity. I think the Boring parts can actively help us to recharge, in moderation.
More importantly, practicing solving Boring problems is how we eventually learn to solve the Interesting ones.
Imagine playing a video game for the first time, and you have a cheat code that lets you skip to the final boss. You'll probably have a hard time defeating it without having learnt the combat system first. Maybe you would actually beat the game quicker this way if you persevered, but I don't think it'd be anywhere near as fun.
Maybe you can optimize for only having Boring problems (much like picking a Boring tech stack), but I would argue that this optimization process is itself non-boring.
Let's accept the premise and say that AI will handle the Boring problems while we humans get to deal with the Interesting problems. How do humans graduate to that level of problem-solving ability, if not through getting practice solving the Boring problems?
Are these limitations inherent?
I think it's an inherent limitation to the way we're currently building and deploying AI. But it might not stay that way forever.
In theory, with enough compute, you could emulate a human brain on a molecular level. But we don't know how to do that, and I don't think we will any time soon. I didn't think we could do it 10 years ago, and although the recent advancements in LLMs are impressive, I don't think they put us any closer to that possibility.
P.S. I wrote 95% of this article in July and then forgot about it. On revisiting it I found it hadn't aged much, not yet at least. So here we are.