The phone goes in the bag. A realistic kit for a personal agent you wear.
An earbud with a small camera, a watch as the control hub, a ring, and an agent running on a server you own. What works today and what is still hard.
Krishan's Tech Blog
I asked my watch for a game for my kid, and had it that evening
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.
Ratchet: a code-size budget for your repo that only goes down.
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.
I can talk to my assistant from my wrist now. The first speech error was my own language setting.
A small Wear OS app on the Pixel Watch sends my voice to my personal assistant and reads the reply back. Most of the work was in one late session with coding agents, and most of the bugs were in speech recognition settings and sideloading.
My assistant picked its own memory library. I made it change its mind.
Jarvis went looking for an open-source memory library to replace its own. Its first pick passed the test and turned out to be abandoned. Here is how it ended up on Graphiti, what the public benchmarks do and don't say, and why the real answer will come from replaying my own data.
My kid played a dinosaur game inside my agent's reply. The hard part was keeping it alive.
My personal agent can now answer with a small web page, drawn in a sandboxed frame inside the chat bubble. Getting a game to show up was easy. Keeping it safe, on-theme, and running when the next message arrived was the actual work.
Twenty-one raises, then never up again. The budget got so tight it made workers delete the wrong things.
A one-way ratchet on code size stopped my agent-written codebase growing. Clamping it exactly onto the tree turned out to be the mistake, and the fix is three hundred lines of deliberate slack.
My worker agent finished the job. A 1 MiB buffer ate the answer.
An orchestrator that talks to a human and dispatches long-running workers has an IO problem that is harder than its scheduling problem. Every design mistake I can point at turned out to be an IO mistake.
My agents proved eleven items done in sixty seconds. None of them were.
Two things make agents ship quickly — a work board a human can actually verify, and a gate cheap enough that everybody runs it. Both turned out to be the same problem, and it is a design problem rather than a discipline problem.
1,169 times an agent rewrote itself. Here is what it changed.
A taxonomy of every self-edit an autonomous agent made over seven weeks, with a pie chart, and the one category that ever made the codebase smaller.