HNHacker News
TopNewBestAskShowJobs

mkozlows

3,439 karma · joined March 5, 2014

submissionscomments
mkozlows··on When I reject AI code even if it works
Completely disagree. I think this is one of the big wins of agentic engineering. When you look back at your own completed change and realize that you made it too complicated because your initial abstraction was wrong, you have to debate long and hard about whether it's worth going back and redoing the work -- is the abstraction actually that bad? Would you really get a huge win by changing it, enough to justify spending another day on the task?

But with the agent, you know that the change will be relatively quick and easy, so the bar to tell it to shift approaches is much, much lower.

mkozlows··on When I reject AI code even if it works
Yeah, but I think there's a difference here: If your coworker puts up code that you don't understand quickly, in most environments people give it an approval, as withholding approval is meant to indicate that there's a problem with the code. It's very rare that you'd actually force them to wait to merge until they've explained the code to your satisfaction.

(There are workplaces where that's the norm, I know -- it tends to be a thing with smaller teams with codebases that everyone understands fully, and much less a thing with larger teams where different people have areas of the code they understand more than others.)

With AI code, though, it's _your code_ and you can't give it a lgtm, you actually need to dig at it until you do fully understand it, fully agree with it, and could justify it to a hostile reviewer. It's a different level of rigor.

Not all engineers apply that rigor, though, which becomes a problem.

mkozlows··on When I reject AI code even if it works
You can also just use AI and keep the scale of your changes small rather than refactoring the whole app with a change? This isn't super-weird.
mkozlows··on Claude Opus 4.8
I was hoping that the web UI would be better -- I like Anthropic better than OpenAI from a values perspective and want to use their products, but ChatGPT in thinking mode has been just vastly better than claude.ai.So my fingers were crossed that these changes would bring it up to par.

But trying it out... alas, no. Simple factual questions where ChatGPT would go do a quick search and get the facts and report them back to me, get a "Great question! [totally invented bullshit]" from Claude, even with this new model and thinking set to high. I have to explicitly tell it to search to get it to look up basic facts, rather than it recognizing that it needs to do that, like GPT does.

mkozlows··on Constraint Decay: The Fragility of LLM Agents in Back End Code Generation
5.2 still had a Codex variant, which this doesn't describe using. It also notably is not using the Codex harness -- it does everything with open source harnesses (which obviously are worse). And while it uses two harnesses with its cheap models, it only uses the worse-performing one of those with GPT 5.2 for cost reasons. (They also don't specify effort/thinking level used for GPT 5.2, but given that it performs worse in their baseline testing than obviously non-SOTA models, I'm guessing it wasn't set to anything high.)
mkozlows··on Microsoft reports AI is more expensive than paying human employees
Yeah, they conflate Microsoft's actions (which are not about cost) with a random quote from the "vice president of applied deep learning at Nvidia," who says that compute costs more than people on his team -- but his team isn't using LLMs for software development, they're literally a deep learning team that is burning compute in deep learning development ways.

If people would do even a little bit of math, they'd see that Microsoft can't possibly be paying more for AI than for developers: They have about 80K employees in product development roles. Senior developers probably cost them $400K all-in.

Do they have a $32 billion Claude bill? I suspect they do not.

mkozlows··on GitLab announces workforce reduction and end of their CREDIT values
The big thing on their roadmap is rearchitecting for something that can handle the increased load, though. Like, they're clearly paranoid that if they don't move fast, they're going to be just as busted as Github.
mkozlows··on Vibe coding and agentic engineering are getting closer than I'd like
The comments aren't an LLM thing, they're a Claude thing. Codex doesn't write those gross hyper-verbose comments.
mkozlows··on Removable batteries in smartphones will be mandatory in the EU starting in 2027
https://www.theverge.com/2022/5/21/23079058/apple-self-servi...

79-pound hyper-elaborate repair kit. Expensive for them to send out, but since only two people will ever want them to, probably amortizes well.

mkozlows··on Uber torches 2026 AI budget on Claude Code in four months
Everything in this article is purely fake. The numbers don't add up, don't match any reported info, and are just fiction.
mkozlows··on Uber Torches 2026 AI Budget on Claude Code in Four Months
This terrible unsourced article seems to be citing this information piece: https://www.theinformation.com/newsletters/applied-ai/uber-c...

... but the key fact about "$500-$2000" per engineer does not appear there, and seems to be fabricated.

mkozlows··on The End of Eleventy
Most Kickstarters have a fake low goal so that they can hit it and "blow past it by 1000%!!!" If a Kickstarter hits its goal, but then still cancels, that typically wasn't their real goal.
mkozlows··on Neovim 0.12.0
I just run them in separate terminals. The only real gap was that I couldn't tell the robot to open files in nvim when I wanted to look at them, the way it could in other IDEs, so I whipped up a quick skill (https://github.com/mkozlows/nvim-skill) to do that.
mkozlows··on Ask HN: How do you deal with people who trust LLMs?
ChatGPT thinking models are very good; the instant model is bad. Gemini is always desperate to find an answer, and will give you one no matter what.
mkozlows··on Meta will shut down VR Horizon Worlds access June 15
It sold terribly. The update was super-minimal, and mostly seemed to have been made for production-simplification reasons (as in: it was cheaper to update it than to keep making the old product, and they apparently didn't want to just cancel it entirely).

Rumors of future products are never super-reliable, but point to their ambitions being downscaled at best. Really, everyone expects them to pivot to smart glasses, because that's what they clearly wanted to make all along, and there's probably a market for smart glasses in a way there isn't for... whatever the AVP was supposed to be.

mkozlows··on OpenAI agrees with Dept. of War to deploy models in their classified network
If you believe that any country should have a military and intelligence apparatus, the job of that apparatus is to kill people and surveil foreigners. I do think the US government should have a military and intelligence apparatus. Therefore, any company that works with it, from suppliers of clothing and food to suppliers of compute and AI, are supporting an organization with that mission.

The US military _does not_ need to build autonomous weapon systems and _should not_ surveil US citizens broadly.

mkozlows··on Statement on the comments from Secretary of War Pete Hegseth
I think it's definitely true that you should never count on a company to do principled things forever. But that doesn't mean that nothing is real or good.

Like Google's support for the open web: They very sincerely did support it, they did a lot of good things for it. And then later, they decided that they didn't care as much. It was wrong to put your faith in them forever, but also wrong to treat that earlier sincerity as lies.

In this case, Anthropic was doing a good thing, and they got punished for it, and if you agree with their stand, you should take their side.

mkozlows··on OpenAI agrees with Dept. of War to deploy models in their classified network
So there are two possibilities here:

1. There's no substantive change. Hegseth/Trump just wanted to punish Anthropic for standing up to them, even if it didn't get them anything else today -- establishing a chilling effect for the future has some value for them in this case, after all. And OpenAI was willing to help them do that, despite earlier claiming that they stood behind Anthropic's decisions.

2. There is a substantive change. Despite Altman's words, they have a tacit understanding that OpenAI won't really enforce those terms, or that they'll allow them to be modified some time in the future when attention has moved on elsewhere.

Either way, it makes Altman look slimy, and OpenAI has aligned with Trump against Anthropic in a place where Anthropic made a correct principled stand. It's been clear for a while that Anthropic has more ethics than OpenAI, but this is more naked than any previous example.

mkozlows··on I am directing the Department of War to designate Anthropic a supply-chain risk
If the government doesn't want to sign a deal on Anthropic's terms, they can just not sign the deal. Abusing their powers to try to kill Anthropic's ability to do business with other companies is 10000% bullshit.
mkozlows··on Layoffs at Block
You can say yes to more of them, _but they still need to be worth it._ If you have an infinite well of ideas to grow your business, awesome. But there are a lot of companies that just don't have those ideas, where their growth is limited on some other factor related to the market they're in.
mkozlows··on Layoffs at Block
Depends on the company, but I think it's fair to say that not every company has a roadmap to infinite growth.
mkozlows··on This time is different
The personal computer, laptops, web browsers, cell phones, smartphones, AJAX/DHTML, digital cameras, SSDs, WiFi, LCD displays, LED lightbulbs. At some point, all of these things were "overhyped" and "didn't live up to the promise." And then they did.
mkozlows··on A beginner's guide to split keyboards
Lots of options, but the one I use and would recommend is the Iris: https://keeb.io/products/iris-se-kit

Advantages over the Ergodox:

1. No pointless layer of "inner" keys that you never use 2. The thumb keys are closer to the main keyboard, so more of them are in a natural reach, rather than being a big stretch (this is the biggest one in usage) 3. Uses all 1u keys, so greater keycap compatibility (any ortho kit will work) 4. If you're comparing to the Ergodox EZ, construction is better, with a metal case instead of plastic 5. Takes up less desk space

And it's still QMK, still hotswappable, still has the columnar layout. I don't think the Ergodox offers anything over it.

mkozlows··on A beginner's guide to split keyboards
I feel like the Ergodox was traditionally a lot of people's first wacky split keyboard -- but for those that stuck around, it was rarely their last. These days, there are so many better options that it's hard to recommend Ergodox despite its historic importance.
mkozlows··on A beginner's guide to split keyboards
That's true for the "Path 1" keyboards this article talks about. The other ones definitely take some time.
mkozlows··on A beginner's guide to split keyboards
Conversely, I'm totally sold on it. The shoulder-hunching thing is so real.

The thing I thought was ergonomic BS was the benefits of columnar/ortho layouts; everyone talked about how your fingers just moved vertically and it was so much better for them, and I rolled my eyes. But dang if it hasn't proved to be meaningfully true for me, too -- when I have to type on a legacy keyboard, I can clearly feel the pain in my fingers. (The disclaimer here is that my fingers are totally screwed up; if you don't feel pain normally, this probably matters less.)

mkozlows··on AI is destroying open source, and it's not even good yet
The wild thing is, that "plateau" link is from September 2025, aka two months before Opus 4.5.

Yeah, it's not a plateau.

mkozlows··on Beginning fully autonomous operations with the 6th-generation Waymo driver
I feel like the bigger issue is that Cruise evidently had an unsafe company culture (like Uber): It wasn't just that they had an incident, it's that they lied about the incident and tried to cover things up.

This has been a pretty consistent pattern -- Cruise was always less transparent about its safety data than Waymo, and its claims tended to be opaque and non-measurable, whereas Waymo was partnering with insurance companies to get hard data.

Waymo is going to have incidents, too, but I think they have made the (correct) decision that being open and transparent about safety stuff is the way they move past those; Cruise made a decision in the opposite direction, and it killed them.

mkozlows··on I started programming when I was 7. I'm 50 now and the thing I loved has changed
I agree with that for programming, but not for writing. The stylistic tics are obtrusive and annoying, and make for bad writing. I think I'm sympathetic to the argument this piece is making, but I couldn't make myself slog through the LinkedIn-bot prose.
mkozlows··on Nvidia's 10-year effort to make the Shield TV the most updated Android device
No. The Apple TV _service_ does, and you can configure that service to be some kind of weird god service if you want. But you can also treat that service like any other normal service, one that only comes up if you launch it. In that case, the home screen is just a straight icon grid with no kerfuffle.
← PreviousPage 2 of 18Next →