airstrike

This demo feels about as good as eating a spoonful of tiramisu

they will likely suffer enormous real losses too, not just paper, though not as enormous

for VCs, breaking even is losing

it's pretty great for coding because the CLI provides the model with all the context it needs and with a way to verify the result of its actions via linting, compiler messages and program output

the rest of "work" doesn't have that luxury

No, compiz was awesome

Models need richer context to do well. They also generally need a verifier, but better context is already an improvement. So to fix it, you need to rethink UIs to be agent-first.

I'm not saying Kimi Work and its peers are entirely useless. I'm just saying they're a hack that puts a pretty low upper limit on agent capabilities and speed.

Says who.

Says everyone who's not using this stuff. It's an empirical argument. Is anyone actually using this for work? How long has it been since Cowork was released?

It's such an inadequate tool for work it reeks of SF hubris to think you can just slap a chat input box onto a PC and replace 90% of white collar jobs.

That's precisely the problem.

People are building UI/UX for a paradigm that is so utterly new that they miss the mark entirely.

Everyone defaults to a big chat input because that's all they know. It's the expectation, so it's safer, faster, and cheaper to not deviate.

It's also wrong and bad for work.

I'm open minded, I just haven't seen evidence to the contrary. Speaking as someone who's spent an ungodly amount of time in claude code

Still a chat interface, so definitely not an "AI Desktop".

Also this was released about a month ago

https://tech.yahoo.com/ai/meta-ai/articles/moonshot-ais-kimi...

yeah, I'm surprised I had to scroll this far to see this

everyone else seems to be thinking the quote is about AI-assisted coding but I read it as the model being used within the product itself

Hmm if you assume the 80% is uniformly distributed between successful and unsuccessful startups/companies, which by default you should, then yeah, your proposed statement is true.

Data would be needed to argue the 80% skews unsuccessful

I started my post with "it seems to me" precisely because I haven't decided that I know best.

OK, this actually worked as advertised. I feel dopamined.

I just wished the delivery took longer and there was a payment process that longer than an instant. Could have a saved card on file that you could use.

This workflow alone is immensely powerful, and takes alot of burden off the team.

ITSM those unsupervised workflows are essentially an attempt at purported productivity in the near term at the expense of meaningful incremental long term burden for teams.

The only ostensible benefit is in the eyes of the AI-psychotic tinkerer, who knows no better, or in those of the clout-chasing developer farming likes on their LinkedIn posts.

This isn't really a bell curve.

Ironically, yes.

it is 100% the argument being made, but you can't see it just yet

idk about you but I'd rather not be murdered regardless of who the perpetrator might be

you should probably seek mental help and read them this specific thread to cut to the chase instead of paying for 10 sessions

Terribly, but just because any two people suck at basketball doesn't mean a third person must suck too.

Well, yes, I have in fact always dreamed of owning an abandoned house in Flint, MI

and famously doesn't require a degree

It's also really hard to buy studios and hope to catch future lightning in the proverbial bottle.

Past performance is no guarantee of future success.

It's a stretch rather than a point.

Hear, hear! One of the best comments I've read in over a decade on this site

I used claude code, codex, and pi. Both claude code and codex feel like hot garbage. Codex sometimes hits 100% cpu after I close the terminal I was using it in and stays there until killed. Claude code will say things like 'press escape to cancel this dialog' but when I press escape it leaves the dialog open and interrupts claude instead. The behaviour of both changes from day to day.

Pi works. I haven't used it heavily enough to have opinions about the design, but it feels like a regular piece of software instead of a fever dream with unit tests. All three are heavily vibe-coded, so I'm curious what the pi folks are doing differently to maintain some baseline level of code quality.

sadly, Anthropic has mistaken the quality of their model for the quality of their tools, so they force their subscribers to stay with their inferior client lest they wish to pay API prices, which I don't understand why anyone would

HN auto edits some clickbaity/tropey titles in submissions.

Submitters have a grace period of several minutes to edit them back to anything they want, which will not get autocorrected, but many fail don't. They may not realize it or be too lazy to care, we'll never know. (Personally, I think anyone who considers themselves a "hacker" would at least try to bypass such a primitive system, but I digress...)

Some HNers complain every time that happens, gleefully oblivious to the countless other times the autoedit feature works as advertised.

It's 2026. CPU goes brrr. It's absolutely trivial compared to the query execution time.

And Django makes it ridiculously easy to write those raw queries in SQL directly so it seems like you're getting lots of mileage from the ORM without giving up anything

The difficult parts are just literally a raw SQL string so how is that any harder?