Five years after the first line of code, Stalwart v1.0 is here: the first stable release, with a finalised database schema, months of profiling and stress testing, a rewritten search engine, and benchmark results that put it ahead of Postfix and Dovecot in 35 of 38 tests.
What I appreciate about this is the acknowledgement of something called “nuance” that AI can now write viable code and be a tool that improves software, reduces bugs and security issues, and increases the speed with which it is written and fixed. Just because there’s an AGENTS.md doesn’t make code slop. And the more we refine this method, the cheaper it gets in terms of tokens/energy.
I’m not happy about what it means for juniors coming into the field, but we might have to change what we teach in order to build good software, as has been done dozens of times before.
We’ll still have the kneejerk reactionaries that are locking in the KV pair of AI=Slop but i guess that’s a constant of the human condition. Conservatives never change their minds.
And there is a last reason, which is less rigorous but matters to us just as much. We write software because we enjoy writing software […] We are not interested in optimizing away the reason we do this.
Holy fucking shit. I read the whole thing and I even enjoyed reading it. This has to be the most sane shit I’ve read all month. This is THE definition of “how to use a tool”. I freaking love it.
Definitely agree on that part, if only the tools themselves weren’t so unethically created and a continual heavy drain on so many aspects of society I might consider using them for small, tedious tasks outside of work where I’m forced to.
Also, they don’t do much to aid my work and actively slow me down quite often since I don’t think in words, so I have to translate my logic into a conversational essay for the AI to understand and then debug code someone else wrote. It’s so much faster for me to translate my logic into programming languages directly which follow logic, not conversational norms. I see how it’s useful for people who can’t think in logical steps, though.
Here’s their AI policy: https://support.stalw.art/t/how-we-use-ai-at-stalwart/1496
What I appreciate about this is the acknowledgement of something called “nuance” that AI can now write viable code and be a tool that improves software, reduces bugs and security issues, and increases the speed with which it is written and fixed. Just because there’s an AGENTS.md doesn’t make code slop. And the more we refine this method, the cheaper it gets in terms of tokens/energy.
I’m not happy about what it means for juniors coming into the field, but we might have to change what we teach in order to build good software, as has been done dozens of times before.
We’ll still have the kneejerk reactionaries that are locking in the KV pair of
AI=Slopbut i guess that’s a constant of the human condition. Conservatives never change their minds.Holy fucking shit. I read the whole thing and I even enjoyed reading it. This has to be the most sane shit I’ve read all month. This is THE definition of “how to use a tool”. I freaking love it.
It was kinda long so i just used ai to summarize it /j
but seriously, wow, it is exactly what I’ve been thinking would be the perfect policy, they really nailed it.
Definitely agree on that part, if only the tools themselves weren’t so unethically created and a continual heavy drain on so many aspects of society I might consider using them for small, tedious tasks outside of work where I’m forced to.
Also, they don’t do much to aid my work and actively slow me down quite often since I don’t think in words, so I have to translate my logic into a conversational essay for the AI to understand and then debug code someone else wrote. It’s so much faster for me to translate my logic into programming languages directly which follow logic, not conversational norms. I see how it’s useful for people who can’t think in logical steps, though.
Yeah it’s quite disappointing how often that part is conveniently ignored in
The AI corporations are hiring Lemmy snipers now?
Those 4 points for AI code writing are extremely well thought out. I can really respect that.
That reads much too sane for today’s
idiot mud fightdiscussions about AI.