… aaaaand we’re back, ladies and gentlemen! That was a nice three week break and I’ve thought quite a bit about what to do with this newsletter when I come back to it.
The result: I’m going to double down on having fun with it. I think I’ve gotten into a bit of a rut, after trying to follow the recipe every week for years.
I don’t know what that means exactly, but hey, you’ve been warned.
We released Agents in Orbs this week. This has been a long time in the making and now it’s finally out and people are already starting to say “in an orb” as if it’s a common phrase, I love it. I truly believe that remote agent— excuse me: agents in orbs, will play a big role in the future. Why? One reason is that these models are incredible when thrown into a sandbo— I’m sorry: an orb. I mean, look at this. There aren’t any magic strings being pulled behind the curtain. If there’s ffmpeg, a model will find a way. These agents need less and less handholding and that includes the handholding by a bespoke development setup. They’re productive in these remote machines. The second reason is that something changes when you can start many agents in many different orbs in parallel. I tried to articulate that in that post up there but based on the conversations I had in response, I think it’s something you really have to try for yourself. But I can add that the more I use agents in orbs, the more I believe that thinking of remote agents as “agents that I can remote control on a different machine that’s similar to my local machine” is the wrong way to look at it. The fact that the orbs are ephemeral changes what you do and how you do it. Just like switching from a single build server to a build system with VMs changes things. State is no longer an issue. Resources and runtime is no longer an issue. These agents in orbs now look like async functions to me, less like remote controlled agents. Async is the point. I now often end prompts with “… and now run all the tests, fix all the bugs you run into, then push” and then switch to another agent. It’s very, very interesting and exciting to try to get them to do more and more in orbs and see how it changes your interaction with agents.
I also wrote about how we made our codebase work with agents and how we made it work with agents in orbs so that they can do a full end-to-end runthrough of our core workflow inside an orb and then present a screenshot: Putting an Agent in an Orb.
At Amp we want Freedom of Intelligence.
The results of the Twenty Ninth International Obfuscated C Code Contest are in and interesting as always. Take a look at the Hacker News comments. They contain some gems. This one here, for example, a comment by the author of the “GameBoy emulator’s code [that] also looks like the GameBoy”: “I first wrote a full Gameboy emulator in C. It started out at about 6000 non white space characters. I then spent about about 100 hours work trying to get it to fit into the 2503 limit. For a long time I wasn’t sure it was going to fit.”
Long and very good post about Turbopuffer: Inside the fastest-growing Canadian AI startup you’ve never heard of. More of this! The form already exists, of course: new startup, short profile of founders, how they grew, how they blew up, etc. But this feels more nuanced and deeper. Good stuff.
Ethan Mollick on working with Mythos: “Last year I called this working with a wizard: you chant the spell and something happens. With Fable the spell has gotten powerful enough that I am no longer sure I am the wizard. I am closer to a patron. I describe what I want, I pay for it, and I judge the result. The conjuring happens somewhere I cannot watch, in hundreds of small choices I never get a vote on. The work has shifted from process to outcome. I no longer steer; I commission.”
Very thorough post on agents and how models have been trained to be agents: Agents need Work Data. Yours truly makes an appearance in a quote.
The End of Determinism. I like the phrase The End of Determinism.
Definitely not in the Joy column, but it’s a good post: LLMs are eroding my software engineering career and I do not know what to do.
19min talk by Tyler Cowen that neatly summarizes a lot of things I’ve heard him say or write in other forms: AI will improve our economy, but will we let it?
Tim Ferriss on whether AI has already killed how-to nonfiction: “My position—and I’d genuinely love to be wrong—is yes, prescriptive nonfiction is the canary in the coal mine, and the coal mine is enormous. I believe LLMs become the interface to everything: search and purchasing, obviously, but also surfing video, summarizing podcasts, navigating courses, even browsing books. The original content doesn’t exactly disappear; it just becomes raw material that most people never touch directly.” But the post is deeper than this quote might suggest. Highly recommend reading it. It’s great.
Very, very, very, very good: Why I Stopped Arguing With People. Read this and if you find it even slightly interesting or even irritating, I recommend reading How to Win Friends and Influence People. Being right isn’t valuable most of the time.
Giant Banana Pulled Over in Montana. I wish I had a giant banana car.
Valve announced the launch of the Steam Machine and here’s what I wrote in our internal Slack: “I’m 100% serious with this: This is incredible writing. This is perfect. This is how to talk to technical customers.” And after that I wrote ten more lines. I also had a fever at the time, but I think that’s unrelated. It’s fantastic writing.
“Something that keeps me up at night is the amount of miscompiled software running in the wild. There’s a famous story, ‘the Core 59 problem’ from Facebook where seemingly random files were missing in one of their Spark databases. After herculean levels of debugging, they narrowed the problem down to a single worker box, on a single CPU core, that was literally doing math wrong. The initial bug-reproducer was 430k(!) lines of code. Eventually they managed to create a 60-line snippet of assembly that reproduced the issue 100% of the time. […]” And now it’ll keep us awake at night too.
“Elderflower is an experimental, open source, independent single-user desktop
OS. Linux kernel, musl libc, no existing distribution underneath. The userland is TypeScript. Apps are written in TypeScript and WASM, no native binaries. The system shell is a TypeScript REPL. All system APIs (filesystem, network, audio, app UI toolkit) are TypeScript.” Watch the video here.
Voytek Pitula’s Fintech Engineering Handbook. Lovely resource.
“During the first two centuries of the Roman Empire, there spread a practice known as hypomnēmata, a type of notetaking system, used as a tool for meditation, in which the writer would store quotes from books they had read. Each day, often in the morning, the notetaker would open their notebook and look for a passage relevant to something they were struggling with, and then they would meditate on that—unpacking it, making the idea top of mind, ensuring it was alive in them.” I’d love to do this, so I set myself the goal to collect these quotes (or videos, or tweets, or articles) and then realized: wait, that’s this newsletter, isn’t it?
This was fantastic and I wish it had been twice as long: Stop eating Lady Gaga’s Oreos. It’s very good. I just went through it again, trying to pick out a single quote to convince you to read it, but it’s very hard. Go read it, please.
The Original Computer Art? Lissajous Figures Created with Oscilloscopes
Nabeel S. Qureshi’s Principles: “A cursed fact of the world is that the most important life lessons you learn are the hardest to communicate to others. They always sound like clichés. In any case, these are a few things I’ve learned from experience and that I try and keep in mind.” Great.
As someone who also loves performance optimizations, I enjoyed this very much: When Impressive Performance Gains Do Not Matter.
Here’s a very nice pairing for you. First, watch this interview with F1 world champion Nico Rosberg. It’s very honest and deep and reflective. Fascinating to hear Rosberg talk about enduring (or not) the pressure that F1 drivers are under, how he handled (or didn’t) the expectations, how even when you’re one of the world’s best race car drivers you don’t think “life is amazing!” but instead you might only feel anxiety and loneliness and not good enough. It’s great and it made me wish I could talk to Rosberg and ask him some things.
And then watch this interview with Tom Brady. It’s a fascinating conversation, both because I find what they talk about interesting (how to set a high bar in a team, “do your fucking job”, “you work Monday to Saturday really, really hard, so Sunday is easy”) but also because Brady is incredibly eloquent and the meta-questions of how and why he answers specific questions in certain ways are interesting. But it pairs nicely with the Rosberg interview, because Brady seems to be the complete opposite of Rosberg. Brady is incredibly confident, his mental game is his advantage; the guy sounds like he never once doubted whether he can be the greatest of all time. Rosberg on the other hand, who also became a world champion, talks about how he’s very defensive, often has doubts, struggles with anxiety, and so on. And yet he also was world champion.
After 9 years, I changed my avatar. Strange feeling. I’ve followed many, many people over multiple decades on different platforms and some of them I can’t picture any other way than their avatar. They are their avatar in my mind. But I felt I was kinda cheating by not weaing glasses in mine (I started wearing them in the year the original avatar photo was taken), so I recreated it and here we are. Also: wow, so many platform where one uploads their avatar.
