A small workshop for people building with AI agents.
You bring something you're actually working on. Others give you the feedback you don't get on Twitter. You give it back.
What this is
An invite-only group of traders, developers, and tinkerers who use AI agents in their daily work — not as a research project, not as a content topic, but as the thing on their desk.
We talk about what's working. What broke. What we wish we'd known three weeks ago. The conversations are concrete because the work is concrete.
Less hype. More receipts. Smaller than you'd expect, on purpose.
Lately in the workshop
- Aprreallyunintented shipped
claude-memory-kit. Three small levers (shortcut rule, writing-memory skill, non-blockingPreToolUsehook) to stop Claude Code's auto-memory from drifting. v0.1 came out the day Claude unpromptedly recalled six-month-old context from an unrelated machine. - AprA rate-limit workaround made the rounds. Route planning through NVIDIA Nemotron and codegen through StepFun Step 3.5 Flash, via VS Code + Kilocode. Posted as “for anyone hitting the Claude limits.”
- AprSix drafts narrowed to two. One of us asked the group to pick a winner from six design drafts. Got told six was four too many. Two side-by-side became the rule, the design tightened.
How to get in
Joining isn't a form. Send a DM with something concrete you're working on — a repo, a workflow, a setup, a recent build. If it fits, we invite. If not, we'll tell you why.
This isn't snobbery. The group lives on members showing up with something to discuss, so the way you come in is the way you'll be inside.
DM @openagentcrew on X →