Maxime Petazzoni

This is a glimpse of big changes ahead of us. If you’re betting on big central models you should think twice. I run the exact same setup (M5 MacBook, qwen3.6-27B, pi, ollama) and while its not as fast or good as one of the big central models, it’s past the line of “cool demo” into “truly useful.” Kind of where the big frontier models were in late 2025. In ~24 months we might have local models that are fast and good enough for most tasks.

The scariest finding in this paper: the subjects couldn't tell it was happening. UPenn ran this study on 48 healthy adults. One group slept 8 hours. Another slept 6. Another slept 4. For 14 straight days. They tested cognitive performance every 2 hours from 7:30am to 11:30pm. The 6-hour group's reaction times, working memory, and sustained attention deteriorated on a near-linear curve. By day 14 they were performing at the same level as someone who hadn't slept at all in 48 hours. The 4-hour group hit that threshold by day 6. Here's the part that should unsettle everyone who thinks they "do fine" on 6 hours: the subjects' self-reported sleepiness flatlined after the first few days. Their brains kept getting worse. Their perception of how impaired they were stopped updating. The cognitive decline was invisible to the person experiencing it. The researchers found a hard threshold. Any wakefulness beyond 15.84 hours in a day produces cumulative neurobiological cost. That cost compounds every single day you exceed it and does not reset with a weekend of sleeping in. About 35% of American adults sleep less than 7 hours a night. 40% of those get 6 hours or less. In 1942 that number was 11%. We built an entire professional culture around a sleep schedule that this paper says is functionally equivalent to pulling consecutive all-nighters. "I'm fine on 6 hours" is the most common response to sleep research. The first thing chronic sleep debt destroys is your ability to notice chronic sleep debt.

My idea to fix F1 is to just race Miatas the rest of the year

🇺🇸 Nearly 2,000 US citizens were asked to point out Iran on a map. These were the results:

It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow. Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes. As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now. It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.

The most effective people I've met all maintain a gigantic locus of control and bias towards internalizing blame. Whenever something doesn't go their way, they deeply analyze the circumstances to identify what they should have done differently, or, even if they acted appropriately based on all the information available, how could they have gotten more information or taken intermediate actions to extract more signal to inform later bigger actions. They don't dwell or make excuses. They just learn and apply that learning to the future so they don't make the same mistake twice.