Storing isn't remembering
There is a lot more to remembering than having the memory stored.
// A personal blog — my own experiments and views, not my employer’s.
Experiments in AI, security, memory & more.
Agent M Dashboard online
System Status:Active
Memories:6.5K
Connections:2.5K
Words:3.1M
Sources:9.9K
Projects:6
Local Time:––:––:––
Illustrative only, based on a real example.
On the tools I build and what I have learned running them.
New here? Start with → Green squares · Experiment: Drive -> Git
Experiment: Drive -> Git
Agent M's memory is at its simplest a folder hierarchy of markdown files. This experiment tested how to store them, simply, but also reliably across devices.
Becoming the assistant
Agent M started as something I installed in a repo, and it kept growing, so it's moving to the host level and ... slowly ... becoming the assistant I had in mind when I started.
The agent couldn't click Allow
When something gets stuck for days, it’s worth asking what it was waiting for.
Google Drive sync has no undo
I learned that “it syncs” is not “it’s backed up” — and I made plans to move the memory under git, to finally get history, diffs, and an undo.
Running Agent M all day
Running an agent all day used to mean hitting the usage ceiling by mid-afternoon. Here's how I cut a heavy day ~90% and made Agent M cheap enough to just leave running.
The multi-surface wall
I learned that “works everywhere” has an asterisk right now: the agent travels, but the surfaces underneath are uneven, and the portable, do-anything agent is getting closer, but probably still a harness generation away.
Building Agent M
I set out to rebuild a trading engine. What I built instead was Agent M — an agent with a durable memory that works across every tool I use.
A 1M context window isn't free
I learned that more context isn’t free — sometimes you want less of it.
Rebuild, not refactor
Sherwood came together fast, but it had drifted too much and needed a re-write. Re-writing it was where I realized that I needed to rebuild something else.
Antigravity -> Claude
So far it’s working well and we’ll see if I don’t end up switching to something else in a few months as the landscape changes.
Don't get too far ahead of the agent
Whether today’s agents can actually keep drift in check without that human in the loop is still an open question.
Building Sherwood
My first agent-assisted project was Sherwood, which I started to learn how markets work. It also taught me some important lessons about agentic coding.
Green squares
Agentic coding took my GitHub contributions graph from gray to green.
Before it had a name
How a one-week work demo became a real cross-platform agent
Intro -> Next Level
AI helped my team build in a week what would have taken six weeks. I decided to write about what building with these tools is like.