@evacchi

Programming Language Nerd, PhD in CS — Principal ML Engineer @RedHat — prev: (Chicory, wazero) @Dylibso @Tetrateio @RedHat — @PapersWeLoveMI organizer

Italy, Milan
Joined June 2010
am I the only one or the the agents are amazing at finding subtle bugs in PRs, but they NEVER suggest to rewrite the patch even when the code is an unreadable mess
2
1
280
I had agents creating breaking changes in surface API and the review agent would not flag it
1
61
I also had agents "eyeballing" very subtle bugs but not even mentioning some unreadable combinator chains with nested closures
45
I guess SKILL.MD issue on my end but today I tried a few LLMs with a real rust codebase and I understood why DHH says he doesn't ever want to look at it
1
1
140
maybe I was too vague: it's because the agents are shit at it
27
Edoardo Vacchi retweeted
I guess SKILL.MD issue on my end but today I tried a few LLMs with a real rust codebase and I understood why DHH says he doesn't ever want to look at it
1
1
140
Edoardo Vacchi retweeted
2
3
137
2
3
137
Edoardo Vacchi retweeted
I see your VirtIO devices you clone 20000x/hour and jam into namespaces across nodes just to read a file, And raise you a light, elegant, virtual stack that pops and pushes tiny ops, with code and data respectfully separate, which only, by design, is cognizant of the simplest of imports and exports, possibly fleeted to a fraction of the overall footprint of the above to accomplish the same fidelity. Yes, we can dramatically improve the Unix process model, by orders of magnitude, in terms of resource use, speed, and safety, while keeping the exact same semantics. People think some of these things are so sacrosanct there is nothing to improve there, when it is quite the opposite. (blog post on above dropping Friday, hopefully 🤞) boxer.dev / blog – at San Francisco, CA
1
17
764
Edoardo Vacchi retweeted
so we really are accelerating 100x because I thought it would take them weeks, instead in the last 3 days people have realized that their “enhanced productivity” really is burnout-speedrunning
Humans, notoriously bad at multi-tasking: “AI makes me so productive: I can do 100x more things in the same time I used to do just one! It’s so exhausting! I’m mentally drained at the end of my day!” That’s nice, have you thought of just doing the one thing in 1/100th the time?
2
1
5
776
My cofounder is @intellijidea
Heard a software engineer say: "my cofounder is Claude" about a project they built + shipped. I am baffled by people humanizing AI, given it's just a tool. The same way my keyboard is not my cofounder (though I could not write @Pragmatic_Eng without it!) but a means to an end
1
6
249
Edoardo Vacchi retweeted
Today I am releasing KillSwitch: a new programming language designed to be difficult for LLMs to write, edit and understand. It makes for a great benchmark suite for LLM models and is a lot of fun to program in. Check it out 👇 killswitch-lang.org
1
3
11
498
Edoardo Vacchi retweeted
I'm looking for my next role, in Montreal or remote! My background spans compilers, systems, and AI research. I'd welcome introductions, leads, or a repost to help spread the word ❤️ More details about my experience and what I'm looking for: pointersgonewild.com/2026-09…
20
118
9
599
99,864
maybe if hollywood fast-tracks a christopher nolan blockbuster about the labs they'll backtrack a little from the whole doomsday schtick wdyt
106
I don't get all this psychosis about RSI I've been using a vertical mouse for years now
1
139
well the prompt is definitely shorter than one page
Sam Altman says Trump and Xi would win the Nobel Peace Prize for a one-page AI agreement “I think Presidents Trump and Xi would get the Nobel Peace Prize together if they could agree on something that should be easy to agree to. And it would be wonderful.” “I think that clearly the two countries are going to compete in lots of ways, and this is going to be important socioeconomically, geopolitically. But they should be able to agree that no one should be taking a certain level of risk with the development process of this.” “And even if just the US and China could agree on some shared standards and testing for development of this technology, I think that'd be a wonderful accomplishment that the two of them can deliver.” “I don’t think this is hard. This is like a one page document.”
1
219
this weekend on the news is one plot twist after the other absolute cinema almost as good as sama being fired
85
Edoardo Vacchi retweeted
Correct take but also extremely tech-bubbly take Software is not dead. The market of selling to tech people is (mostly) dead There are billions to be made in the next few years by solving normie software problems for normies A plumbing business will not be vibe coding PlumbCRM
I just saw this @ycombinator's batch literally only has harness startups or hardware startups So the proof is in the pudding In the most early adopting tech pioneering place that's Silicon Valley new startups aren't even building software anymore And it's debatable if anyone will actually need a custom harness or it won't just be generically offered by the AI frontier companies So software is mostly dead and hardware it is
53
21
6
506
50,118
Edoardo Vacchi retweeted
What we used to mockingly call "GPT wrappers", are now respectable "domain-specific harnesses".
I was just at YC demo day yesterday. Besides hardware/physical things, everyone is just basically just building a domain-specific harness
49
221
11
6,071
159,885
imo a lot of people are deeply misreading this as lamenting the death of personal pleasures in the face of efficient goal-attaining when it's a lament of the death of actual goal-attaining in the face of optimized things-shaped-like-goals-attaining
the terrance tao crashout is something to behold
30
393
22
3,522
87,022