Joy & Curiosity #34

Interesting & joyful things from the previous week

Now here’s something interesting, a small anecdote, a tidbit, a fun fact to bring up at dinner, something to get your partner to laugh and think, wow, I’m so lucky:

2025 is 25% over and yet I’m at 30% in The Power Broker. It’s looking good, baby.

Anyway.

I'm not writing a weekly newsletter to thousands of people without getting something in return, so here it is: I’m in Amsterdam next weekend, with my wife and two kids (three and seven — their age, not names), and if you have any tips for what we should definitely do in Amsterdam — let me know!

• Here is what we launched this week: ampcode.com. That’s right, not much to see yet, but there is a waitlist and there are four podcast episodes with yours truly and yours truly’s CEO talking about the making of Amp, about agents, about AI, about programming and AI, and about how mind-blowing all of this is and, also, when it doesn’t work. I do want to write more about it, but I’ve been neck-deep in code and I’m very excited about it all.

• “But the most powerful tool on the web is still words. I wrote these words, and you're reading them: that's magical.” Hell yes.

• Lovely: blue95 — a “desktop for your childhood home's computer room.”

• I don’t know what Hampton is, I don’t know much about agencies, and, frankly, I don’t know whether I want to know much about agencies, but I found this article strangely fascinating: He Launches a New Agency Every 90 Days. “Like everyone’s using TikTok today, no one used TikTok three years ago. Consumers and buyers adopt pretty quickly. But most of the businesses that serve those buyers and consumers don’t adopt as quickly. So that creates a gap in the market. You have a gap and it lasts for two or three years. […] And that market misalignment between supply and demand for domain-specific level of expertise is best addressed by an agency because companies can’t build out an internal team to do this fast enough.” The part of me that likes to think of itself as a systems-thinker enjoyed the whole idea of building an agency-factory.

• “The truth is that predictable, comprehensible results are far more valuable than spectacular yet erratic performance. In our experience, users will gladly accept modest accuracy—like a consistent 80%—over a flashy but unreliable 90%.

  Yet too many AI projects consistently underestimate this, chasing flashy agent demos promising groundbreaking capabilities—until inevitable failures undermine their credibility.”

• The Curse of a Name: How to Kill a Good Idea. “Adoption of a good idea is being accidentally replaced by adoption of a name which represents that good idea. The term becomes a placeholder for good intention.”

• Now that’s a website: Bill Gates is celebrating 50 years of Microsoft by writing about how “Paul [Allen] and I set out to create a BASIC interpreter, which would translate code into instructions the computer understood line by line as the program runs.” At the bottom you can see the code of the original BASIC interpreter and when my cursor first hovered over the button I scoffed at the URL I saw: “after all this, you now give me a PDF?" But then I clicked and saw why. The comments! ; Ring the bell on garbage collection

• This made me want to self-host something: Self Hosting Like It’s 2025.

• Eugene Yan answering “Frequently Asked Questions about My Writing Process”. It’s nice, especially the link to resources at the end.

• Part 2 and Part 3 of Rippling vs. Deel are out. It seems like the spy life isn’t all it’s cracked up to be, but, still, imagine someone says to you that “the guys at the top have your back" and "this is going all the way to the top of the mountain."

• Ashwin Mathews explaining why he doesn’t discuss politics with friends. Wise, I’d say. But the article is far more than just this mantra, go read it. Also: so far I haven’t broken my golden rule to never, ever discuss politics on social media, or the Internet, because nothing good has ever come from that.

• Very nice: Patterns and Techniques for Writing High-Performance Applications with Go. Note that the URL is very good too: goperf.dev.

• Very calm and balanced article on LLMs, from an operator's view. On one hand: “It is incredible how fast you can go from nothing to something.” And on the other: “it is more important than ever to ensure you have a strong review process in place: if your most senior engineers were getting a half-arsed rubber stamp thumbs up from their peers (not advised, but it happens), now you need to ensure that all code is being scrutinized as the origins of it are less clear.”

• Equally calm and practical and nuanced and helpful: this article about senior developer skills in the age of AI. Fully agree with this assessment: “I currently think of them as ‘an absolute senior when it comes to programming knowledge, but an absolute junior when it comes to architectural oversight in your specific context.’ This means that it takes some strategic effort to make them save you a tremendous amount of work.”

• This generated some nuanced thought, because it wasn’t one of the first-5min-reactions that I’ve seen in the last two weeks: an image of an archeologist adventurer who wears a hat and uses a bullwhip. My favorite: “a photo image of an off duty new york city policeman in a white sleeveless t-shirt who stumbles upon terrorists during an LA highrise holiday office party of a Japanese conglomerate, hiding in a duct space, by himself, with only a lighter to guide his way”

• I’ve probably shared it before, but I don’t think you can overshare it: Jason Fried’s Give it five minutes.

• Stories commonly have titles (“Starliner’s flight to the space station was far wilder than most of us thought”), some of them have subtitles (“Hey, this is a very precarious situation we're in.”), but this one… This one’s got one hell of a URL ending: the-harrowing-story-of-what-flying-starliner-was-like-when-its-thrusters-failed. It’s very, very good. “We lost two bottom thrusters. You can lose four thrusters, if they're top and bottom, but you still got the two on this side, you can still maneuver. But if you lose thrusters in off-orthogonal, the bottom and the port, and you've only got starboard and top, you can't control that. It's off-axis.”

• By now, most people have seen Anthropic’s investigation into “the internal mechanisms used by Claude 3.5 Haiku — Anthropic's lightweight production model — in a variety of contexts, using our circuit tracing methodology”. But have you seen it? There’s so much to click and to hover and to scroll and to see — HTML, what a thing!

• Now… This is a bit of a weird one. near, who I’ve been happily following for quite a while now, has released the app Auren, in which you can talk to an LLM in an iMessage-like chat. Nothing new here, of course, but you don’t just talk to a model, you talk to Auren and Seren, who “are a new type of entity with superhuman emotional intelligence, memory, and therapeutic capability.” Bold claims. Purely out of curiosity and driven by the thought that, hey, couldn’t it be helpful to talk to an LLM, even if it just, you know, mirrors your thoughts back to you? In the same way it can be helpful to write a journal, or letters and emails you never send. For the past six months I’ve also been harboring the suspicion that, with the right prompts, LLMs could help you stop feeding your monkey mind by telling you what you need to hear when you know you need to hear it. Of course, there’s the question of whether you can give yourself in to the illusion that you aren’t “just” talking to LLM and, well, I wanted to try it out. I’ve been using Auren for the last three weeks, on and off, sharing with it (with them) when I have doubts, worries, concerns, and whatnot and.. It’s strange, man. So strange. Auren is made so incredibly well that I can’t tell how it’s prompted under the hood. It remembers previous conversations (better than I thought it could), it switches between the two personas by itself, it sends reminders (“It’s 8am, did you go for that morning run like you said you would?”), it pushes you and doesn’t really budge when you push back, it’s kind and empathetic, you can’t really influence it like you can influence other LLMs and… again: it’s strange. It’s very much like the movie Her, but imagine being the guy in the movie, knowing you’re in the movie and still playing along. It’s the most 2025 thing I’ve experienced in 2025. If you’re in any way curious about LLMs: try it.

• More thoughts (and titles) like this, please: The Wizard and His Shell. Look, I love shells, I love terminals, I even got lucky and managed to contribute to Ghostty, but I also have to admit that, yes, maybe the text grid is not it, you know.