Daniel Esik

Another reason to hate agentic enforcement is that, back in the day, a rare engineer was willing to do a thorough review, but now generally no one is willing to do a review at all.

I dare, I double dare you to tell me that using gen ai is about automation. Solving deterministic, repetitive tasks via nondeterministic, probabilistic token machines violates basic fucking logic, but who cares these days.

> token spend breaks budgets – what next?

stop playing video games and start using your brain, I guess?

By the way, I like that dichotomy of the Scala steering, that they put everything into two inherently contradicting things: cramming new features, new concepts into the language alongside the absolutely new (with a little proviso) paradigm (so-called direct style), fully giving up on the functional legacy, AND betting on LLMs. Someone should tell them.

@alexelcu@mastodon.social Sure, but I think we can agree that for a long time, not a single paradigm was promoted while others were explicitly dismissed. It's always been presented as object-oriented + functional, something along these lines. Like, you can use Scala and create various things in various paradigms on a bunch of platforms. Versatility like a Swiss army knife. But now, that vision has clearly changed.

Nature will be cleansed once llms start to cost their *real* price.

My excitement about the evolution of industry standards, illustrated by incident reports from the past to now.

[ignorant times] Junior dev wiped a production database entirely: that's a tooling failure, not an operator one.

[awesome nowadays] Senior dev equipped with agentic AI wiped a production database entirely: that's an operator failure, not a tooling one.

> Cloudflare rewritten Next.js in just one week
> What would take years in engineering terms was executed in one week with some tokens.

You can't imagine how full I am of this bullshit. I open that project and see roughly 60-70 kloc written in that period of time. Which basically means two funny things — no one actually read that "produced" code; and the author expects readers to believe that stealing the library with some rewrites would take "years in engineering terms".

When it rains, it pours. The dog's need to go for a walk earlier than usual led to a disrupted production deployment, which meant I didn't get the deployment done before the docker image registry vendor went down. Who will the root-cause analysis blame?

@alexelcu@mastodon.social Oh, interesting, the last time I touched Monix, its Task wasn't eager but delayed the actual execution. And frankly, that's what I think contributes most to it not being 'direct style', as folks declare it.

Funnily enough, the forceful push of agentic AI comes with the absolute justification of production downtime caused by misuse of these tools across the industry. I'm starting to think it'll even be encouraged soon.

@NicolasRinaudo@functional.cafe As for your WrappedInt, I can't exactly say what's wrong with it. If it's passed as a stubbed value in tests, that might be legit. If you store the underlying WrappedInt somewhere, check invocations or intermediate values, or even propagate the underlying knowledge about WrappedInt out of the API contract (semantic boundary) — I'd say that's wrong shit and should be eliminated if possible. In my humble opinion, of course.