– Yerba Mate carries me through my mornings. After reading Atomic Habits, I’ve been more observant of routines that are positive towards my productivity. And I’ve landed on this morning productivity routine that makes doing chores, writing on this blog, and of…
Agentic coding gave everything in my apps an undo button: layouts, features, translations, bugs. Everything except the data. That's where I slow down now.
During my last holiday I spent a good chunk of its 2 weeks working on a new side project (more about it in a future post!). A lot has changed with coding agents recently, and it has been a while since I’ve had to push their limits on a relatively complex probl…
The play phone
The play phoneHaving a son has made me much more conscious of how I spend my time. Not just because there's less of it that's entirely mine, but because there's now someone watching what I do with...
The death of web development education
Rescuing a field from disappearing
I Ran a 397-Billion-Parameter Model on 20 GPUs With 16 GB Each
Splitting Qwen3.5-397B across 20 rented GPUs that can only reach each other through a relay: what limits the speed, what two days of testing fixed, and the real numbers.
The tiny TPU started as a matrix multiply. Once a compiler could target it, the programs showed where the time actually goes.
Gemini 4 Argon: our next era of frontier intelligence
Announcing Gemini 4 Argon, our frontier model for real-world coding, enterprise knowledge work, and cyber defense, rolling out soon.
I visited the Museum of the City of New York today and got to see He Built This City: Joe Macken’s Model, the 50 x27 feet model of the city …
Read the History, Not Just the Code
Hi everyone 👋 When you join a new project, what do you actually do? You open the code. You read it. And pretty quickly you hit something strange. A weird conditional. A function that takes a parameter nobody passes. A workaround with a comment that says "do …
A few sentences of prompting and then hours of <Enter> pressing, what a wonderful day (no)! YOLO it for the god's sake.
We Will Forget How to Code. And that's OK
AI may change which programming skills we retain and value. As agents write more code, developers will need to judge its architecture, correctness and failures, and beginners will need ways to build that judgment.
I’ve been in a funk for a few weeks and am taking a couple of weeks off work to reflect on the past and rethink my future. Much of this has been driven by how AI has changed software engineering. I spend most days chatting with Claude. I occasionally chat with…
My great-great-grandfather thought he’d be replaced by technology too. He was a farrier in Mandres-en-Barrois, the small village in rural France where I grew up.
What an LLM server does when ten people send a prompt at once: prefill, decode, the KV cache, continuous batching, PagedAttention, and more, with animations.
Letters To My Kids
I’ve been writing letters to my kids for over a decade and will soon be handing them over.
Building an Ultra-High Throughput AI-SQL Engine
Quail jointly plans AI-SQL queries and model inference. Across 29 QUAIL-B queries, it is 1.84x faster on average than well-tuned vLLM baselines.
Pining for Arc Downcasting in Rust
In the course of writing my build driver, I came across a bit of an unusual problem, for which I made a bit of an usual solution. I think...
My last post on the Rust compiler’s performance was two months ago and a lot has happened since then.
This post describes the construction of an Aho-Corasick automaton for the simultaneous matching of substrings within a sequence. I’m fond of this algorithm because it constructs an automaton from an existing tree data structure in a rather pleasant way.
The AI Race Just Got Awkward | insufferable.dev
Funny how quiet everyone got.
Pi now supports MCP. Why we changed our minds, what changed in MCP and in Pi, and what Codemode is.
Photo Scrubber — local face blur & metadata removal
How a Pole sees the world in September 2026: a scrolling essay that follows one closed strait into fuel, food, bonds, defence and demography.
GitHub - ninjahawk/livenerf: Benchmark for tracking model capability after release.
Benchmark for tracking model capability after release. - ninjahawk/livenerf
GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price