When The Future Doesn’t Need Us
What does the world look like when AI outcompetes humans?
What does the world look like when AI outcompetes humans?
On prompting, specificity, and learning the words.
Last week we released version 0.2 of pgrust. This release was all about performance. It’s 10x faster than the previous version of pgrust. On OLTP benchmarks, pgrust is 30% faster than Postgres, and on Clickbench, Clickhouse’s benchmark for analytical databases…
Our Moon is two-faced. Science missions unravel its layers.
Moon Monday #286: China’s Chang’e 6 samples have helped scientists decipher enigmatic layers of the lunar farside, leaving a curious trail for future missions to build upon.
Kimi K3 built a playable first-person game from one prompt, but its bugs only appeared when a human played it. AI can make the first version. Somebody still owns the failures.
Richmond loses more construction workdays in January than in July, on a third less rain. Why spread-out winter rain costs more than clustered summer storms, three wrong guesses about a model that read too high, and what happened when it was measured against a …
Your first company can start with a problem you already feel, and the tools you need while solving it.
I am quite tired of Algorithm in any app, so am I with Spotify.
Retries are great for transient failures, but they can't solve eventual consistency. Treating one problem as the other often adds unnecessary complexity to distributed systems.
As AI steadily reshapes the software industry around it as open source once did before it, it’s useful to try and understand where those two forces intersect. Open source has, as it has with virtually every other industry software category, had an enormous imp…
We once argued that swarm theory and promise theory were the right lens for resilient systems. Then we built Multisynapse to prove it. Here's what shipped, and where we're going.
I Built a Digital Human You Can Talk To
An engineering account of a voice-to-avatar pipeline built with Deepgram, Groq, Chroma, AWS Polly, GeneFace++, FastAPI, and WebRTC.
Everyone is panicking that AI will bury us under incomprehensible bugs. Relax. The code is about to get so perfect there will be no bugs left at all, only the business cases nobody documented, or documented badly. The machine builds exactly what you ask for, w…
Edward Teller once said, “Von Neumann would carry on a conversation with my 3-year-old son, and the two of them would talk as equals…” He added that he sometimes wondered whether von Neumann used the same principle when speaking to everyone else. That is a rem…
The compression scheme lost to a small model trained from scratch. That loss led to the real result: at a fixed time budget, a 5M model beats one 28 times its size, until it doesn't.
Practical guide to shipping production-ready software in hours instead of months.
I spent three years inside the most ambitious company I've ever seen, and the way it works is a preview of how every company will soon have to work.
Working on Economics with Fable 5 - Wilsons Blog
For the past few months I’ve been working on a theory. It started out as just a fun little data exercise looking at some different types of taxes and benefits and how it effects what people buy and how much they work. During that time I took advantage of opus …
My newsletter has no subscription form, no double opt-in with a magic link, and it doesn't live in a third-party service with its tracking pixel. And
Agentic coding collapses the cost of implementation labor. Why the rational bet is moving from high-level libraries back to primitives you own.
I have not used a corporate VPN in more than ten years. At Google, access to internal applications depends on the user and the device, not on the…
In this post I'll talk about how my team and I designed a local multiplayer investigation game, how I implemented a record-replay mechanic, the strange platform bugs we hit along the way, and how we still managed to make a game we're proud of, using my engine …
When testing our eBPF agent, I don’t always get the same experience as our users, especially in performance critical sections. I realize that the benchmark test suite isn’t always enough, because user’s environments can be completely different from our benchma…
Honey, I shrunk the embeddings: Matryoshka vs. PCA – Dylan Castillo
I don't write code anymore. I approve it. An honest look at the slow shift from author to approver, what AI took from the craft, and what was already fading on its own.
Reinventing the generic iterator wheel.
An interactive exploration of how outfielders (and dogs) catch fly balls: the gaze heuristic, optical acceleration, and why a ball hit directly at you is the hardest one in the sport to read.
The age of personalized software is here.
A company’s public statement in the middle of an outbreak is not marketing. It is evidence — what the company said it knew, and when, published to the
Login
August 3, 2026
August 3, 2026
August 3, 2026
Login
August 3, 2026
August 3, 2026
The JFrog security research team recently identified a supply chain attack targeting the `xinference` package on PyPI. Versions 2.6.0, 2.6.1, and 2.6.2 were compromised and yanked by maintainers after users reported suspicious behavior. If you installed or imp…
April 16, 2026
August 3, 2026
August 3, 2026
Too often I ask a question in Slack or leave feedback under a merge/pull request or argue with friends in a WhatsApp group and get back:
I just wanted to do an internship in Europe so it would be easier to find a job after graduation. That internship completely changed my life.
During the pandemic, we completed one of our largest projects at work. To celebrate, since we couldn't meet in person, we all ordered food on DoorDash and played an online escape room game together. W
Qwen Studio
QWEN STUDIO DISCORD Today, we are officially releasing Qwen 3.8-Max, the most capable model in the Qwen family to date. This also marks the first time we will open-source the weights of a Qwen-Max-class model — the open weights will be released next week. Buil…