I’m happy to report that I’ve reached a new milestone in my life: I can now ride my bike with no hands.
For all my life I thought that it’s something incredible hard. I thought of it like I thought of the ability to juggle three balls. It’s something you have to practice and only people with serious dedication end up being able to do it. But then, last week, my friend said he bikes to work without his hands on the bars. Huh. Then I told my wife about it, saying that’s incredible. She said, “why? Everybody can do it, right? I can do it.” What? My wife, who (and let me say: I love her) has managed to bump into every wall of this house, forwards and backwards, just by trying to walk here, can ride her bike with no hands? Maybe… Maybe I can do it?
Yesterday I did it. On a bike ride with my wife, who immediately taunted me by taking her hands off the bars and waving them around, I did it. Took my hands off the handlebars and rode my bike for a few hundred meters. “You look so happy,” she said.
This is one of those posts that immediately make me think: “I’m going to reference this a lot in the future.” I sent the author, James, a note to thank him for writing it and told him: “I recognise a lot of stuff in there that I've seen or done or know to do, but lacked the words for.” I read the book The Nature of Software Development many, many years ago and, to this day, I think it’s one of the best things that have been written about agile (lower-case) software development. And a lot of what Ron Jeffries wrote in his book is echoed in James’ piece here: the “Strategy is a Cone” framing, the stepping stones (obviously), the acceptance of unknown unknowns, the “you can’t just sit there and scratch your chin and figure it out” — highly recommended.
Very important post that I recommend reading: Building AI Products In The Probabilistic Era. It touches on a lot of things that are changing, fundamentally. “Stop for a moment to realize what this means. When building on top of this technology, our products can now succeed in ways we’ve never even imagined, and fail in ways we never intended.” And then there’s this: “With AI products, all this is no longer true. These models are discovered, not engineered. There's some deep unknowability about them that is both powerful and scary. Not even model makers know exactly what their creations can fully do when they train them. It's why ‘vibe’ is such an apt word: faced with this inherent uncertainty, we're left to trust our own gut and intuition when judging what these models are truly capable of.” And here’s something that I’ve felt too but that’s very hard to explain: “This doesn’t work anymore. The more you try to control the model, the more you’ll nerf it, ultimately damaging the product itself. Past a certain point, intelligence and control start becoming opposing needs.” Very good post.
“What if your agent uses a different LM at every turn? We let mini-SWE-agent randomly switch between GPT-5 and Sonnet 4 and it scored higher on SWE-bench than with either model separately.” The era of model alloys is upon us, I think. (My teammate Camden talked to Beyang about this exact topic, in case you’re interested in his and our thinking on the topic.)
SolveIt looks interesting. I found out about it through this video, which I haven’t watched in full, but in the video you can see “Jeremy Howard and Johno Whitaker present SolveIt, a development environment designed to mitigate the downsides of ‘vibe coding’ by encouraging deliberate, step-by-step problem-solving.” And if you click around at this timestamp here you can see that in action. If nothing else, it’s an interesting idea.
“Existing search tools on Windows suck. Even with an SSD, it’s painfully slow. So I built a prototype of Nowgrep. It bypasses most of the slow Windows nonsense, and just parses the raw NTFS.” I had no clue that there’s more greps being worked on. I thought ripgrep won the game and the game’s over. But maybe not on Windows? (Also: I had never heard of BareGrep and just the screenshots alone bring back memories.)
As some of you know, some of my pet interests are espionage and corporate espionage (the whole Deel vs. Rippling thing was my jam, as they say) but also cybercrime with state actors, and so when John Collison asked Brian Armstrong in this episode of the Stripe podcast Cheeky Pint “what does the general tech public not appreciate about the cyber crime landscape?” and Brian Armstrong replied with “there’s a lot of North Korean agents trying to work at these companies” my heart started to beat a little faster. Very interesting 5 minute section in that episode.
This was a ton of fun and made me think (like I did many times before, probably naively) that this is how complex topics should be taught in school: Moving Objects in 3D space.
Do you know who Eoghan McCabe is? He’s the CEO of Intercom. And this, as I found out yesterday, is his personal homepage: eoghanmccabe.com. It’s fantastic on many levels, the styling is just one thing, but look at it: this is truly a personal homepage. There’s a bio, there’s some thoughts, there’s interests, there’s hobbies, there’s photos, there’s links. So good.
Very, very interesting: “Tidewave Web for Rails and Phoenix: a coding agent that runs directly in the browser alongside your web application, in your own development environment, with full page and code context.” I think this is only the start of frameworks and models melting, because, at the end of the day, I think, agents and frameworks try to solve the same thing: reduce the amount of code that has to be written.
Jujutsu For Busy Devs — very, very good. Finally learned about `mine()` being a valid revset (which is exactly what I was looking for earlier today).
“In a manner of speaking, that smaller Rust is the language I fell in love with when I first learned it in 2018. Rust is a lot bigger today, in many ways, and the smaller Rust is just a nostalgic rose-tinted memory. But I think it's worth studying as an example of how well orthogonal features can compose when they're designed as one cohesive whole.” (Side-note: I didn’t know that you could use a hashbang cargo invocation to run Rust programs like that, including TOML and all)
matklad on the TigerBeetle blog: Code Review Can Be Better. Most review tools nowadays I don’t find that interesting anymore (I also did a near 180 on code reviews in the last one and a half years and am now less enthused about it), but this paragraph stood out to me: “When I review code, I like to pull the source branch locally. Then I soft-reset the code to mere base, so that the code looks as if it was written by me. Then I fire up magit, which allows me to effectively navigate both through the diff, and through the actual code. And I even use git staging area to mark files I’ve already reviewed” Now, that’s how you should probably review code. I’ve never done the unstaging before, but a proper, proper review requires checking out the code, I think, and the unstaging/staging is very smart.
Levon Helm: “I don't know. I guess it's from being born in Helena, Arkansas. That's a pretty basic part of America where there's a lot of good basic music. Drums just always sounded like the most fun part of that good music for me. I had the opportunity to see some of the traveling minstrel shows years ago, with the house band, the chorus line, the comedians and singers. In those kinds of shows, with horns and a full rhythm section, the drums always looked like the best seat in the house.”
