I once worked at a company that had a big sales kickoff. The event was at a big hotel, and the entire sales team was staying there too.
On the second day, everyone got sick.
You could see it spread in near-real-time in Slack: "It got me too," "Uh now I feel it," "I won't make it to dinner."
Someone offered to share their medicine: "I have some Tamiflu if someone needs some. I'm in room 302."
An engineer, who wasn't attending the event but followed along in Slack, spotted a once-in-a-lifetime chance and posted: "Tamiflu is for closers only!"
I think he got reprimanded for it, but I'll never forget him.
How fast can you open 1000 files? Daniel Lemire got nerd-sniped by a Jarred Sumner tweet and decided to find out how fast you can open files. The results are surprising: “The macOS system has a faster disk, faster memory and faster cores. Yet opening files is clearly much slower under macOS according to this test.”
Probably the link of the week for me: If your product is Great, it doesn't need to be Good, written by Paul Buchheit, creator of Gmail. He writes: “What's the right approach to new products? Pick three key attributes or features, get those things very, very right, and then forget about everything else.” By now, I agree. It’s the German car door again. Also: “If your product needs "everything" in order to be good, then it's probably not very innovative (though it might be a nice upgrade to an existing product)” Very good, short, succinct — highly recommend you read it.
Another very good piece on products: Avoid the nightmare bicycle, by Geoffrey Litt. “One of the worst misconceptions in product design is that a microwave needs to have a button for every thing you could possibly cook,” he writes. The kicker is this: “Good designs expose systematic structure; they lean on their users’ ability to understand this structure and apply it to new situations.” I think these nine, very short paragraphs finally, finally articulated a lot of my struggles of the past years working on developer products.
Apparently, “BACK IN THE DAY, if you made a New Folder on a Korean computer, the folder was automatically named after a bird (‘new’ in Korean is homonymous w/ ‘bird’). If you made hundreds, the folders would start ‘begging’ through their names, like ‘I've run out of bird names,’ ‘Please stop’” — I don’t know and don’t want to know whether that’s true or not, but, naturally, I asked Claude to generate me an
mkbirshell function so that I, too, can create folders that are named after birds and so that I, too, can see the computer beg after running out of bird names. And Claude did it in less than a minute, just based on the screenshot of that tweet. Here it is. If others say that we’re drowning in “AI slop”, then let me be the one to say: it’s time to let a thousand flowers bloom, man, let the birds chirp.Confetty — it’s “confetti (or fireworks) inside your terminal.”
In case you missed it: I was a guest on the Agents At Work podcast and had a tremendous time. It was not only a very enjoyable conversation (Jordi and I kept talking after the recording ended and only stopped because we had other meetings come up) but I think I also managed to share a lot of the thoughts I have on AI and the future of programming.
This was surprisingly good: 40 Thoughts On Turning 40. Surprising because I assumed it would be filled with clichés, but it really wasn’t. It’s a good list.
Rule #2 of the Mustard Club: “Every Member shall once at least during every meal make the secret sign of the Mustard Club by placing the mustard-pot six inches from his neighbour's plate.”
I found out about the Mustard Club through Robin Sloan’s newsletter and that also lead me to this piece by him: Is it okay? That’s the question he asks about AI models and how they’re trained and where the data comes from. It’s an interesting piece of writing — I can’t say I loved it and I don’t know whether that’s because the questions made me uncomfortable or whether I thought it had too many big words and felt too clever. In any case, I’m glad I read it and wonder now: what does he say about open-source models?
If you haven’t: go install Claude.app and add some MCP servers. I know, I know: MCP is the acronym of the week and it’s very hard to see what’s so important about it. But if you don’t want build an agent yourself and see how magical it is, go and play with some MCP servers. It’s very eye-opening.
Speaking of: Geoffrey Huntley “decompiled” Claude Code.
This paper about a futile Attempt to Catch Up with JIT Compilers was very interesting. Futile? Yes: “We eventually discovered, that the optimized IC sequences are actually no faster than un-optimized ones. The experiments we present in this paper show that in the context of IC sequences, the micro-architectures already executed the extra memory reads of the C sequences early, and there is no benefit in removing them.” And: “None of the tools we tried, including performance analysis tools that use hardware performance counters and micro-architectural code analyzers, was able to predict or even unveil this hardware optimization” Imagine having to write this: “At best, it speeds up programs by 4.0% […]. At worst, it slows them down by 1.8%. On average, there is a 0.03% slowdown.“ The paper makes for a good pairing with Emery Berger’s Performance Matters. Performance is hard. (I do love the “Summary by Example” section in the paper. But I don’t love that they used “prop” as the name for a “property”)
Fabien Sanglard dove into the history of the FastDoom codebase to figure out “why fastDOOM is fast”. Fantastic stuff. If you need a definition of grit: “I wrote another go program to build every single commit. I will pass on the gory details of handling the many build system changes (especially from DOS to Linux). After an hour I had the most ugly program I ever wrote and 3,042
DOOM.EXE. I was pleased to see the build was almost never broken.”I’m writing a lot of TypeScript these days and, guess what, I like it, but I’m constantly confused by the ecosystem and the choices the ecosystem collectively makes. This short, zoomed-out overview of “What is TypeScript?” was refreshing.
Little curveball for you: Joe Delaney, one of my favorite lifting YouTubers, who has in recent years somehow transformed into somewhat of an Eric Cantona of lifting YouTube, with advice after 20 years of lifting. What he says at the timestamp that I linked to here, about “enjoyable is optimal”, took me many years to realize and matches an older draft that I in my notes called “do the thing that makes you do the thing.”
Thomas Wolf, one of the co-founders of Hugging Face, wrote a very honest, very good article on the Einstein AI model. I mean, read this: “I’ve always been a straight-A student. Coming from a small village, I joined the top French engineering school before getting accepted to MIT for PhD. School was always quite easy for me. I could just get where the professor was going, where the exam's creators were taking us and could predict the test questions beforehand.
That’s why, when I eventually became a researcher (more specifically a PhD student), I was completely shocked to discover that I was a pretty average, underwhelming, mediocre researcher. While many colleagues around me had interesting ideas, I was constantly hitting a wall. If something was not written in a book I could not invent it unless it was a rather useless variation of a known theory.” This ain’t no humble bragging. One of the questions Thomas asks is: are AI models just “good students”? “Some of the most recent AI tests are for instance the grandiosely named ‘Humanity's Last Exam’ or ‘Frontier Math’. They consist of very difficult questions –usually written by PhDs– but with clear, closed-end, answers. These are exactly the kinds of exams where I excelled in my field.” I won’t spoil it for you, but I find the last line in the article very fascinating, motivating. (To be honest: but maybe that’s only because I wasn’t a good student?)
Pierre finally launched. You can sign up at pierre.co. You don’t have to, but I need you to click and look at that landing page. Rumour has it that this is one of the most “eye-melting websites every invented” (fictional quote). I’ve played around with it a bit and it’s lovely, it’s different, and I couldn’t have predicted how much joy I’d feel for just a different code host to appear. Let a thousand code hosts bloom (and melt eyes)!
Discovered a cover of The Soul of a New Machine I’ve never seen before and now I want it framed. Some people asked me: do you recommend the book, should I read it? And the answer is: yes! It’s a very, very good book. I love it. Bryan Cantrill wrote about it: “Since reading it over two decades ago, I have recommended The Soul of a Machine at essentially every opportunity, believing that it is a part of computing’s literary foundation – that it should be considered our Odyssey.”
Not a big upgrade-my-OS-because-I-can kinda guy anymore, but I saw that Instruments 16.3 contains a new Processor Trace. And since I love Instruments.app I’m now considering upgrading macOS, just to play around with that Processor Trace.
If Paul says to read I’m here