An agent with its own email, its own accounts and a capped agentic card that needs my tap for every charge can buy things in the real world without ever touching my wallet. Here is how lunch proved it.

My chat has two AI assistants with different personalities. Every message needs a router that decides who answers. Keyword rules got it wrong half the time. A small local model, given the last few turns, got it right more than four times in five, in about 50 ms.

I asked my personal AI assistant to find money I was wasting. It read a year of card charges, caught a duplicate subscription, and talked a support agent down from $129.81 to $86.46 a month. It also made a few mistakes worth telling.

I made an Alexa custom skill that forwards what I say to my own assistant. The code was short. The four bugs were not, and two of them were not in my code at all.

The day after a half marathon, I told Jarvis to order food. Two minutes later, lunch was placed. Here is how goal-aware agents work, where the automation slipped, and why checkout still needs a human.

Analyzing 642 merged pull requests across 16 major open-source repositories: why coding agents bloat codebases by adding logic without pruning dead code, and how code-size ceilings prevent it.

My personal agent can change what my Wear OS watch face looks like from a spoken sentence. A small Java generator draws the face, a headless test checks it, and the watch installs it over the air, with no Play Store, no ADB and no laptop.

An earbud with a small camera, a watch as the hub with the LTE link, an optional ring, and an assistant on a server you own. The full design: device roles, links, tiers of compute, battery budget, privacy, a simulated day and an honest scorecard.

I talk to a personal agent from my phone and my Wear OS watch. I asked it for a game for my kid, then kept asking, and ended the evening with five games on my wrist. They are small and disposable, and I wouldn't have built them any other way.

A small tool that refuses any change which grows your code past its budget, and makes every raise an explicit, logged grant that shows up in the diff. How it works, how I run it on its own repo, and where it does not help.