Classic environmentalists making the acceptable the enemy of the clearly worse.

But in any case underwater data centres make zero sense. If you weren't already convinced by using your brain for 5 seconds, Microsoft proved this a decade or two ago.

If you want easy access to seawater just build it on the coast. There's a reason nuclear power stations are built on the coast and not underwater!

Agreed. I use it for coding too. Even using it as a kind of interactive editor is a huge productivity boost. E.g. I was working on some assertions that were all like assert (... |-> a && b && c) (|-> is implication) and I was like "Claude, split the consequent conjunctions into separate assertions " and it did it in about 5 seconds, vs 10 minutes of tedious hand editing.

Yes indeed, the cheap models still produce slop. Slightly annoying because I'd love to use them for personal projects, but as you say I can't really justify the cost.

For work though it's a no brainer, given my salary and the productivity boost. At my previous company I sometimes spent an entire day debugging a complex test failure that Claude can figure out in 5 minutes. No exaggeration.

Nah just vibe code the stuff you host in Rust. These are all memory safety bugs.

My exception is Home Assistant which is unfortunately Python 🤢 but it's not exposed to the internet.

If you didn’t use a profiler, then how can you be sure that the issue was Python and not something else?

It was a relatively simple program and I am good at programming.

The point is, I put the same amount of effort into writing the same program in Python and Rust, and Rust was 50x faster. With AI, I don't see any reason to use Python these days anyway. I never bought the productivity arguments before (Rust is easily as productive as Python), but with AI Rust is definitely more productive than Python because of the much stronger static checks.

Did you use a profiler to find the bottlenecks?

No. It was a simple enough program that there were no areas for improvement anyway. I also did not use a profiler for the Rust code.

good Python devs know how to find bottlenecks in their software and replace them with fast implementations

In my experience this is also a sweet sounding lie to make people feel ok with using Python. Yes in some case you make a big mistake and there's a large bottleneck you can easily remove... Or your program has a nice small hotspot where it spends most of its time and you can just write that bit in C++ (AI is the obvious example).

But most programs are not like that. They're more like MS Word. Slowness is evenly distributed through the code. There is no hotspot to remove.

As an example of this, look at Nicholas Nethercote's excellent series of blogs on making the Rust compiler faster: https://nnethercote.github.io/

The "bottlenecks" they find result in 1% speedups.

My only caveat: I really haven’t tried Fable due to cost. Anyway, be well, mate. Enjoying the conversation despite disagreeing.

Well, that's why we're disagreeing 😄

I agree with everything you've said if you're using models weaker than Fable. The latest models (Astra, Opus 5.5, maybe Argon I haven't used it much) are a significant step up. Way more than the benchmarks suggest. It really is "ask and you shall receive". You might have to ask a few times but still...

Also caveat on my end: I've only tried vibe coding with Rust, because it's the best language available for humans and AI seems to like it too (lots of static checks). I have no idea the quality of code you'd get in other languages. I would be cautious before having it write C++ for example... though it's probably fine.

If you need any of those things, you can just ask for it.

I get that you really don't want it to be true that it works as well as it does. Your job depends on it.

Mine does too, but I can face the truth even if it is unpleasant.

Also I wouldn't necessarily have believed it either if I hadn't tried it for myself. I would encourage you to spend a bit of money and a couple of days vibe coding something non-trivial with Opus 5.5. Then at least you'll know what's coming...

Counterpoint, I wrote a Python program that reads a load of JSON and runs regex searches over it. It was quite slow. Rewrote it in Rust as a straight 1:1 port, no algorithmic changes. 50x speedup.

I think you must have either done something very wrong or your Python program didn't really do anything other than call Python built-ins that are actually written in C.