hombre_fatal

nvm added like 3+ seconds to my new shell startup, and one of the maintainers defended it with "so what, how often do you open new terminal windows?"

One thing I've learned about tech is that you can never assume something is too boring for opinions.

I keep lazygit open next to each Claude/Codex session that's working on a change, but over time scrolling the changes has gone from necessary to just out of curiosity.

At this point of 2026, if you're reading code and still catching things, then the problem was before the code was written. You should have decided more high level preferences like invariants before the work began.

And this is a good thing.

LLMs have changed the obstacle of doing something from (time, attention, expertise) to (time, attention, tokens).

I don't see what's unique about GTA6. It's just larger in scope if you wanted to decompile or greenfield a custom engine.

But at the end of the day you'll be able to unpack the assets yourself and immediately have the worldmap running in a browser in Three.js. And once you're there, you can vibe-code any game you want on top of it.

> Apple doesn’t seem too happy about agents

I don't get this reaction to Apple making Full Disk Access more explicit. Whether they're "happy" or "sad" about agents doesn't seem responsive at all.

Kinda seems like whenever you spend 10 seconds thinking about the average user, social media gets angry. The quoted justification by Apple seems reasonable.

It's a good habit to always create an upper bound on things (input max length, queue size, etc).

The person who wrote it just came up with 20 in the moment and forgot to go check, something everyone has done a hundred times.

I've probably never worked on a single system where the html validation, http server validation, and database constraint were synchronized on username max length. You choose a placeholder, an even number between 10 and 16, then forget to ever check.

LLMs could write perfect code in 2 seconds every time and you'll still be on the hook for figuring out what makes something fun to play and nice to use. In fact maybe even more on the hook since you wouldn't get to procrastinate by working so much on impl details.

> why not randomly ask a student at the start of each class to present their solution on the board

This makes the class really awful. I had professors with schemes like this, maybe calling on someone's name randomly 1-2 times in class to answer a question they'd just asked, just to ensure everyone is scared into paying attention.

Another data point: hello.c in GTK4 used over 100MB last I checked.

You lambasted the interface (linear messages, chat) but then never pitched us your vision of a successor.

Your technical ideas are just implementation details behind the interface. What's your idea for a better UX?

I mainly use Codex/Sol to review my plans drafted by Fable. But beyond that, Astra blows through usage limits too fast to be a daily driver and writes weird code despite what my "house style" is, and Codex is behind Claude Code in terms of critical features like seeing what's going on in subagents.

The parent + subagent workflow has become critical for keeping the reasoning agent (parent) context-lean while also letting me chat to the main agent while work is getting done.

My main process is to use Fable to reason and then spawn Opus subagents, and I get amazing results, and I'm always looking into what the subagents are doing.

> they will never, ever suggest writing a tool to make their own life easier, get more accurate information, or anything

Make sure you process doesn't depend on anyone reading your mind.

When I run into things like this, it becomes a one-liner in my instructions/harness or in the canned prompt/skill I use that sets off a process.

In this case, I instruct agents to proactively build/improve diagnostic tooling if it would help them with their task + if it meets a bar of generalization/reusability (else it should be an ephemeral probe that gets abandoned at the end of the solution).

Yeah, the biggest example is performance optimizations that sacrifice your data model to the point that you'd never accept them.

I think it's one reason why ADRs are an important of a software project, especially with LLMs. You need a place were you can document invariants, why you have them + the rejected ideas and acceptable risks.

It helps smart agents like Fable help you decide on trade-offs and it's kind of incredible to witness that happening.

Yeah, but that's just the scientific process of hypothesis -> evidence -> conclusion.

You need a measurement that can falsify hypotheses and reject branches that won't work.

Also, if all you have left in your project are performance issues that are hard to identify without flailing around (even with Fable/Astra) despite sampler/profiler reports, then you're doing really well and I wouldn't assume you're going to fare much better than the sota models in terms of stabs in the dark.

"Rank the top findings/solutions by impact vs confidence" continues to be one of my best quickwins to add to all sorts of prompts. Bam, now you have a reasoned priority list.

If it can be measured, then LLMs can optimize it.

Once I had repo commands that could dump `sample` results and a cpu profiler/trace and then a benchmark tool that let me A/A + ABBA/BAAB-test the current modified git workspace against HEAD or any commit, the LLMs could just do their thing.

And that's how my homemade terminal uses much less memory than ghostty/kitty/iterm yet has more throughput.

AI is going to increasingly unmask people and companies who don't care about correct and performant software now that it's become so trivial to guarantee both. It used to at least be expensive and time-consuming and expertise-demanding to do those things.

That $20 plan is going to give you maybe 10min of Astra usage before you hit your limit.

Once you price out similar specs for a Windows gaming build, I don't know how much better you'll be doing, and it won't have unified memory so you'll be "paying twice": once for RAM and again for VRAM.

$250 just for 2x8gb DDR5 RAM. I'm not even sure you'd save much money on a DDR4 build, all parts considered.

Harnesses try to solve a wide range of complex, open-ended UX problems, so there isn't a perfect one.

CC might be the best one I've used if I give each major UX aspect a rating 1-5 and then average them. For example, it has a decent subagent viewer.

Meanwhile, Codex doesn't even have one, and its subagent tool call is so buggy that the parent agent sometimes doesn't even know why the child died.

Being a TUI is very limiting, yes, though that limitation isn't the harness' fault.

I'm generally impressed with the code that Fable/Opus is writing for me these days; I would be proud to have had the same foresight had I implemented the solution myself.

And Fable's architectural design is pretty much always well-reasoned and a good place to start.

There's this idea that the best way to use LLMs is to be in the backseat constantly yelling out corrections, but that hasn't been true in my experience for quite some time, though I only use a few sota models.

I think something being slop this late in the game is mainly a reflection of the person using it. I can't really blame AI anymore when pretty much any lever you'd recommend to de-slop it is one prompt away.

I agree with your last bit, and that is the only thing left now that AI solved the technical part.

That's separate from LLM usage though.

By this point, if you care about a quality product, you should learn how to leverage LLMs for that like running automated audits for correctness and performance opportunities.

There isn't a market for "yeah there's a memory leak but I wrote it by hand."

Agreed. I haven't written a line of code since last November when AI had the big shift into something good.

The average HNer would imagine me scrolling TikTok while I periodically swivel to Claude to type "continue" while my brain shrivels away.

That I can delegate lower level work like implementation means I spend my time at increasingly high leverage positions like deciding what gets built. And any time I want, I can get as granular and inquisitive as I'd like about the lower level work. After all, I probably chose it from a multiple choice AskUserQuestion tool call, each with a blurb and shortlist of trade-offs.

All this talk of de-skilling doesn't make a lot of sense to me because what are these skills people covet so much that defined most of the day to day of a software developer? What, the implementation details of some framework or the intimate details of your company's crappy legacy software stack?

Dunno, I have bigger plans for myself than that.

Javascript's async-everything is really unique in a domain where async is almost always bolted on to synchronous-everything in some sort of incompatible subecosystem.

Sure, but this subset of user was going to otherwise reuse their password and be susceptible to cred-stuffing.

The point is to stop the attack and prevent users from accidentally hosing themselves.

This is pretty cool as a next-gen mini replacement.

The one downside of my iPhone 13 mini is the rare case where I do need more screen.

I couldn't spend $2000 on a phone, but I don't mind the direction.

To know what you're getting. No further reasons needed.

No, I don’t only want personal blogs and content that’s supported by ads. And you should know how to bypass paywalls by now.

The posts that gain traction and hit my feed are filtered to be the best results, but I don't see what that changes. So are the Minecraft posts.

Either way, you'd have to do more work to claim that providers are specifically optimizing for Minecraft vs. other hypotheses like Minecraft just having more resources, preexisting implementations, and base model knowledge than other games.

If we're just going by vibes, I'd wager we're at the point that it's more straightforward to improve the model generally than try to special-case specific demos that are popular on social media.

You can also randomly generate a password for the user on the form they'd normally type one in on registration. Add a "Regen" button to give users more visceral control over it before they submit the form.

Attacks have been distributed for quite some time if your service has any loot worth attacking. You have to handle the case where every request comes from a unique IP address.

Well, I’d say LLMs scale up the underlying starting point.

It’s still shocking to me pre-AI how often people would ship code they didn’t bother to run once, relying on some QA stage to report an issue.

With AI the solution to it is also trivial since an extra three seconds spent writing a prompt could have had the agent run and verify it.