Sean Goedecke suggests reading code in layers: first the overall structure, then the sub-structure, then the details. That reminded me of a 1972 experiment by Bransford and Johnson about how much a title changes what you understand.
I used AI as a consultant for a personal decision this week, until its agreeableness made the conversation go in circles. So I borrowed Architecture Decision Records from software engineering and wrote one for my own life.
I spent a week in my hometown without my AI agent and barely touched social media. It felt like vacation, even without days off. A video on dopamine habits, a flea market and a friend's new hobby made me wonder which skills are still worth mastering.
I want to be transparent about how I use AI when writing. Reading ingredient lists for my gluten-free diet gave me a way to think about it: there's a difference between a post that was AI-generated and one that may contain traces of AI.
I wanted my AI agent to handle listing my old stuff on online marketplaces. Building the skill took so many rounds of checking that I could have listed everything ten times over. An old xkcd chart explains why I stopped.
My screen time said 6.6 hours a week. My chat logs said a whole day. Why the two are so far apart, and why I'm rethinking how I talk to my AI agent.
I want my AI agent to manage appointments for me, but he doesn't get access to my private calendar. Instead, he got his own calendar. I subscribe to it on my phone.
Three weeks after switching from OpenClaw, I dug through my agent's database and found it had searched its long-term memory 9 times across 1,990 messages. The reason: me, not the agent.
It's Wednesday night, I have nothing ready, and my AI agent could mimic my style well enough. Why I didn't let him.
I never set out to hand over my to-do list. It started with a Telegram problem, went through a shared Obsidian vault, and ended with an agent that ranks my tasks against my six-month goals.
11,000 page views in an hour, 48,230 unique visitors in a week, and a burger I could barely eat. Eight weeks after publishing my first post ever.
I built a small tool that turns my RSS feeds into an e-ink newspaper, so I can read my favorite blogs on a pocket e-reader instead of my phone.
Everyone predicts AI will take over writing code. I don't doubt it, but the prediction bores me since it misses the actual innovation. Some French postcards from 1899 show why.
My early experiences running OpenClaw on a Raspberry Pi 4. Covering setup, security, token management, and why I named my little helper Eddie.
Building with AI means someone has to define what 'correct' means. Do it up front, or you'll be stuck checking every result by hand.
2025 was the year of AI agents. But lately I keep reading another term more and more often: the harness. A short explainer of the machinery around the model.
Passwords are annoying. Here's how to make yours strong, easy to remember, and actually motivating.
Multitasking leaves me drained and makes it hard to switch off after work. Here's the obvious solution.
YouTube and Reddit pulled me in for longer than I'd like to admit. Here's the three-step process I used to escape the algorithm.
We all know YouTube and Reddit are useful, but why is it so hard to put them down? In this post I explore what I think is actually going on. These apps work a lot like slot machines, and not by accident.