I'm conflicted on this. I can see the argument, but surely we also live in a world where we want people creating valuable things and making them freely available to be handsomely rewarded and incentivized for doing so. We don’t want people thinking, “You know what? Screw going

x.com/michaelnstruwi… And given the lack of empathy towards the Tailwind folks, he is entirely justified.

You will ultimately be held responsible for the code you commit — if you’re nervous about committing code you don’t understand (such as code generated by an LLM), you are not “making yourself the bottleneck”, “fighting the model”, or “being unproductive”. Being nervous precisely

Oh boy has this piqued my interest. Super fast checkpoint / restore on an entire VM (is it a VM? I don't know enough about Fly.io + Sprites) is exactly the kind maybe-Jevons-Paradox-inducing thing that has me curious. fly.io/blog/code-and-…

Smart insight.

Might just be me... but GPT-5.2-Codex is very bad in Cursor.

I’m an open-minded person — but this is almost certainly due to an over-reliance on shipping fast without understanding the consequences of what CC outputs (since it’s the primary method by which the team developers CC?).

Getting lucky is not the same as good engineering. You’re still sampling from a probability distribution, no matter which way you slice it. This approach seems to do a good job of shaping that probability distribution — but you’ll need to run a decent number of experiments to

This so beautifully summarises what I’ve been feeling but unable to articulate re: AI writing code — what ultimately matters (and will always matter) is the *intention*. Any tool, whether AI or otherwise, is about minimising the gap between the creator’s intention and the

Because sadly, most software engineers hate programming. They weren’t attracted to this career for the craft, only the money. They resent having to write code.

"what do you do?" "I'm an expert on AI ethics" "wow, you must know a lot about AI" "very little, actually"

People are spending so much time making their prompts and agent.md and skills stricter and more deterministic, oblivious to the fact that we already have this unambiguous predictable method of defining programs… called a programming language.

The hardest part was never writing the code itself.

People who care deeply about their craft will anyway be better at it than those who don’t care as much. Not sure if causal (likely), but definitely highly correlated.

The issue with software was never that there was too little of it. I’d love for the mean quality of software to eventually improve with new technology / tooling (LLMs)… but we’re not really seeing that happen yet (or even if it will happen at all). We’re just seeing a massive

I’m grateful someone did the leg work, because this was my suspicion also — the majority of the benefit here is just keeping context clean and tidy. While I was working at OpenBB for example, we learned this very early on — even just on 32k context models. The workaround was

*All* technology is amoral (I.e can be used for good or evil).

Except, very few people have a job that they’re passionate about to this degree, and they’re willing to sacrifice for to this extent. Very few jobs are “great” or even have a possibility of being “great”. There is no incentive or inspiration. Very few white collar jobs will

I’m continuously blown away by how good Sam is at positioning. Just purely from a technical point of view.

If you unironically call yourself a 10x engineer, you’re probably not a 10x engineer.