This week was full of code and ideas and pushing and trying and discovering and throwing away and trying again and delight and joy and ease and hard problems and amazement.
It made me think: the most productive & joyful teams I’ve worked on were similar in some fundamental ways.
They were small (two to three people), barely had any process (code reviews only to get help; never approval), pushed & shipped every hour and every day, constantly talked to each other in unstructured ways, used what they built as much as possible and built plans purely based on what they discovered, made and re-made these plans every day, and put in the work.
And I’m starting to wonder whether everything else — the processes, the structures, the rules — isn’t a distraction. I mean, of course it isn’t, but… maybe it is?
Anthropic released Claude 3.7 and Claude Coder. It already shot to #1 on the Aider leaderboard. And it’s also #4 of the same leaderboard without, to quote Paul Gauthier, creator of Aider, “WITHOUT USING THINKING”. In my personal testing, Claude 3.7 has been mind-blowing. If you’re up for some weekend hacking, here’s something to try: build a tiny CLI application that lets you talk to Claude 3.7, with extended thinking, and give Claude three tools —
list_directory,read_file,run_terminal_command. That’s it. That’s 200-300 lines of code. Then you run it and ask it to help you improve the program you just wrote.Also: did you ever notice Anthropic’s 404 page?
Mind-blowing how spot-on the title is: crossing the uncanny valley of conversational voice. I just spent four minutes talking to Maya, forgetting that it’s Maya for ten seconds at a time.
Imagine waking up and someone tells you: “You are a JS and opensource-software specialist, going to explain some uglified JS code.”
Half of the Internet already linked to this video of someone running DOOM as TypeScript types and the other half has already commented on it saying how insane it is (it is!), so let me just add: not just the project, but the video is really, really good. It’s also 6m59s, go watch it.
“Overall, the feeling is like those videos of Neil Armstrong on the moon. He’s bounding. Programming in the normal way feels like walking. You type out each expression, stepping incrementally toward your goal. When programming with AI, each move is bigger than a step. You lift off the ground. It requires more forethought, but, because you make more progress with each move, it feels like flying.”
Benedict Evans, ruthless: “OpenAI and all the other foundation model labs have no moat or defensibility except access to capital, they don’t have product-market fit outside of coding and marketing, and they don’t really have products either, just text boxes - and APIs for other people to build products.” Another stand-out line is this one: “This reminds me of an observation from a few years ago that LLMs are good at the things that computers are bad at, and bad at the things that computers are good at.“ I call this the fuzzy-to-non-fuzzy adapter.
I love that John Siracusa is still out there, shipping applications for the Mac. If you’ve never read one of the Siracusa reviews of macOS, take a few minutes and read through this bit on memory compression in the review of OS X 10.9 Mavericks.
Here’s a pot of gold for you. It’s summaries of discussions between “Robert ‘Uncle Bob’ Martin and John Ousterhout between September 2024 and February 2025”. It’s Clean Code vs. A Philosophy of Software Design. It’s actual code being discussed. It’s Ousterhout asking: “Bob, can you confirm that you stand by this code (i.e. the code properly exemplifies the design philosophy of Clean Code and this is the way you believe the code should appear if it were used in production)?” And Martin saying “Ah, yes” and then Ousterhout saying “there are many design problems” with the code and that it’s “chopped up so much (8 teeny-tiny methods) that it's difficult to read” and adding “those names are problematic even for someone who understands the algorithm” — for pages and pages and pages. I’ve only read half of it. But this is fantastic. This is real. They really, really talk about how to write code.
So, DeepSeek apparently wrote their own “high-performance distributed file system designed to address the challenges of AI training and inference workloads” and, yes, that’s sick, but: the name, man. The name is fantastic: Fire-Flyer File System. Say that out loud. Say it again. Right? If I were one of the people working on it, I’d never say 3FS. It’s Fire-Flyer File System every time, all day long.
Developing Taste: “If you are a designer, you should be looking at great designs. If you are a writer, you should be reading great books. Expose yourself to great work, this way you'll learn how greatness looks and feels like.”
macOS Tips & Tricks — I thought I knew macOS, but apparently I don’t. Going to try to use some of these.
A glimpse into graphics programming and floating point numbers: A Shader Trick. Here’s a spoiler that isn’t a spoiler: “So the reason the fog is snapping after you play for a while is that the time sent to the shader system gets reset to 0 every 1000 seconds but the fog doesn't know about this. And the reason for the resetting is to make sure we maintain good precision in the time-based effects in the shaders.”
Long Ben Taub piece on the photographer Paolo Pellegrin. It’s about a master photographer becoming blind. It’s about subtraction (“But now I find myself—let’s call it in the full maturity of my life—and I find myself doing the opposite, and that is that of subtraction”). It’s about mastery (“I have always found it mesmerizing to observe someone who is among the best at something, doing it as it ought to be done.”) and beauty (“How do you render an experience of the sublime? You address the idea of infinity”). It’s great.
Shoutout to Readwise Reader. Exceptionally well-made app. The focus on power-user features feels very rare nowadays. Love it.
This post by Sarah Catanzaro was great: Annotation for AI doesn’t Scale. “Instead of, for example, identifying trucks on a highway, now we have annotation tasks requiring domain expertise […] how do you decide which massive blob of code is better? […]. But as the coding tasks get longer and harder, it becomes harder for humans to catch bugs and other problems.” While and ever since doing fine-tuning last year, I keep thinking: this is fucking hard, really hard. Then I watched this talk that also says: this is really hard. (Click on the link — it’s timestamped — and watch 2-5min, to get a glimpse). Now Sarah’s saying it. (Also, side note, the “original paper on RLHF” linked in Sarah’s blog post is from 2022! 2022! I thought it was at least 10 years older.)
One of my pet topics is cyberwarfare and “state actors” and cyber spies and, well, this post about a “a North Korean Phishing Operation Targeting DevOps Employees” checked a lot of my boxes. Using a hijacked LinkedIn profile they tried to recruit someone for a DevOps role, lured them in, and gave them a coding exercise in a Bitbucket repository. But that contained a credential-stealing backdoor. Talk about the job market these days, huh.
Fantastic Morgan Housel piece. Very, very, very good.
In last week’s issue I shared my love for single-use websites such as everytimezone.com. Love came back. Mariano shared the following: strftime.org, flexbox.malven.co, grid.malven.co, and NoTengoEnie.com to easily copy the ñ character if you don’t have it on your keyboard. Last week my daughter asked me how to type an ö on the ANSI keyboard layout of the Linux laptop and instead of explaining to her how “alt-right is the compose key and then you just—”, instead I googled “oe” to get “ö” to show up. Maybe I should build IchHabKeinOe.de.
I know, I know, I know, this sounds a bit fortune cookie, but I believe in this cookie: “Opportunities unfold as you execute. Not as you whiteboard. […] Everything changes the moment you start. New information surfaces. Constraints shift. Hidden doors appear that were invisible from the sidelines. […] Execution is the unlock.“ I actually found myself in a discussion this week, before I came across the linked post, saying: “you can’t fix this by sitting at home, thinking, you have to do something, anything, to start figuring it out.”
Taking a short break from The Power Broker, I read this tiny book on creativity by John Cleese. 112 pages. No big revelations inside, but it is charming and tiny and full of good reminders.
gopls now has a modernize tool. Interesting. Is this a shift in what language servers do? (You could take the comments above each analysis function and feed them to an LLM and it would also modernize the code, I bet.)
You don’t even have to click, just let this quote simmer: “Two AI agents on a phone call realize they’re both AI and switch to a superior audio signal ggwave”
Here’s Edward Z. Yang on being on parental leave and using “Cursor on an unfamiliar project”. I spent the past two weeks essentially doing nothing but coding, using Claude and a growing an agent, and can confirm how important it is to be “careful what you tell the model to do”, that they’re “fucking nuts at UI code” (I’m new to Svelte and Claude helped me actually create some really nice UI components). People laughed at the Prompt Engineering title but I think knowing what to say and which context to add is still 90% of the game.
Jon Krakauer’s one of my favorite writers, but I didn’t know there’s a battle waging on YouTube in which one side’s trying to pull Krakauer’s Into Thin Air (one of my favorite books) down into the dirt. Until I read this article.
What I try to do with nearly everything: “ritualizing ordinary moments makes them sacred. and when something becomes sacred, when you give it meaning, it gives meaning back to you.”
Through this article I realized what a wonderful word Wunderkammer is. Cabinet of Curiosity doesn’t come close. Wunderkammer — wonder chamber, wonder room, except not quite. “Often they would contain a mix of fact and fiction and were collected from exploring expeditions and trading voyages. Honestly, it didn’t matter though, in spirit these cabinets were not meant to be scientifically accurate. Those who could afford to create and maintain them, could construct for themselves their own version of the world.“ What would my programming Wunderkammer contain? Or is this it?
Somehow I ended up reading this whole speech by John Gardner, called Personal Renewal. I copied out quite a few passages; too many to share here. Instead, here’s two to convince you to read this. “I'm not talking about anything as narrow as ambition. After all, ambition eventually wears out and probably should. But you can keep your zest until the day you die. If I may offer you a simple maxim, ‘Be interesting,’ Everyone wants to be interesting — but the vitalizing thing is to be interested. Keep a sense of curiosity. Discover new things. Care. Risk failure. Reach out.” Maybe I should chisel that into the wall of this newsletter? Or maybe this one here: "The future is not shaped by people who don't really believe in the future. Men and women of vitality have always been prepared to bet their futures, even their lives, on ventures of unknown outcome. If they had all looked before they leaped, we would still be crouched in caves sketching animal pictures on the wall"
