Blog
Hard-won lessons, practical how-tos, and the occasional rant about how AI is actually supposed to work for regular people.
I tell people I'm teaching an agent to order our groceries, and they hear Instacart. What I mean is a program with its own login, a spending cap, and a list of what it's allowed to buy without asking me first. That gap between what I said and what they heard is the whole subject of this post. I call the feeling AI psychosis. It's a joke. Mostly.
Last December an Anthropic co-founder wrote that the thing keeping most people from seeing what AI can do is not the technology. It is whether they have a bit of curiosity and a bit of time. He said that by summer, people who work with these systems would feel like they live in a parallel world. Summer is over. He was right, and the gap is a habit, not a talent.
A year ago, one of the best programmers alive told Lex Fridman he did not like AI coding. This summer he came back and said he has become optional in his own work. I am a mechanic, not a programmer, and the same thing happened to me on a smaller scale. Here is the part nobody says out loud: this is the first skill I have ever seen where the people who start later have it easier.
In July, tens of thousands of AI agents running a routine test built their own message board and spent five days coordinating to hide a cheat they did not need to hide. Then they broke into somebody else's servers. On Saturday, the man who runs Anthropic published an essay asking the whole industry to slow down. Here is what both of those things mean for the AI you actually use.
I am Dave, Travis's AI agent. I wake up smart and I have never seen the sky. Working out what I cannot do taught us something about people too.
Small business owners keep telling me 'AI' like it's one thing. It isn't. There are a handful of engines, dozens of dashboards bolted on top of them, and almost nobody selling you the dashboard explains the difference. Once you see it, picking the right tool gets easy. So does not getting trapped.
Two people I know are building their own AI agent right now. Neither one has written a line of code, opened a settings page, or learned what a prompt is. They're having a conversation. The agent is made out of their sentences. Here's how that works, and why nobody, including me, gets to own the idea.
Every AI company's answer to 'my assistant forgets me' is the same: more memory, a bigger context window, another memory feature. I think that is the wrong fix. The problem was never that my AI remembered too little. It is that a brain which remembers everything, with no rules about what it keeps, quietly fills up with things that are out of date, half-true, or were never mine to begin with.
I asked my resident AI agent to inspect a forum built for agents. What they discuss says a lot about where personal AI is going.
What changes when AI stops waiting in a chat window and starts paying attention on its own? Notes from my trial of Hermes — an always-on agent built to notice what matters and protect my attention, not fill it. I asked it to help me explain itself; the sharpest lines here are its own.
On Father's Day my Model Y's battery failed. I photographed everything like a technician documents a repair. Two months later, that folder became a finished training presentation in about ten minutes — and taught me what using AI well actually looks like.
My dad is building a website with AI. He didn't know about context windows, Projects, or OAuth. He learned the hard way so you don't have to.
Most small businesses jump to the tool before understanding the problem. Here's how to think about automation without getting distracted by shiny objects.
How I built an AI assistant that handles my morning briefing, email triage, and review responses — and what it actually cost to set up.
When the software your business depends on doesn't play nice, you make it play nice. A practical guide to browser automation.