~/community
Agents, in public. Barcelona, and on GitHub.
I build with AI agents on production work and I write down what holds. This page is the community side of that: a meetup, the repos, and the rooms I show up in.
~/next
Agents in production. Barcelona.
A small meetup for people running agents on real systems, not a demo night. The first one is aimed at mid November. The date and the room are not set, so this is a waitlist, not a ticket.
placeholder · email until the event page is up
How I use Claude Code
most days
Claude Code is the daily driver. The skills that stuck are the unglamorous ones: an incident, a queue, a merge request, a release note. Each one ends in a diagnosis, evidence someone else can check, and a command that is printed and not run.
Read by default. A person approves anything that would change production. The long version is the note, and the three that followed it: how a team learns from mistakes, why the write stays in a separate step, and which model should do which job.
The read-only note →Close the loop →Read by default →Models by role →
Open source
the shape, not my team's skills
agent-ops ↗
Read-only Claude Code skills and a shared standards shape, for people who run things in production. MIT. It is the scaffolding, so another team can lift the shape without copying anyone's stack.
Claude Code skills
How a skill should end: a diagnosis, the evidence, and the command left for a person. The article is the why. The repo is the starting files.
Events
rooms I went to, and the one I am starting
Agents in production, Barcelona. First one aimed at mid November. Waitlist open, date not set.
Join the waitlist →AI meetup in Barcelona, on the 5th. I went expecting models and agents and spent the night with founders. The note is on X.
The post ↗
Want to talk at the next one?
A short talk from someone who has run an agent on a real system, including the part that failed. Tell me what you would show.