Training
Building an AI Software Factory In preparation
Eight labs with Claude Code
Eight hands-on labs that turn a repository only its author can build into a small software factory, with Claude Code as the agent: runs you can start, check, cost and trust without reading every diff.
Pre-registering is an email to me, nothing more. I write back when the first sessions are scheduled. No newsletter.
At a glance
- Who it is for
- Developers and tech leads who already use a coding agent every day and want to go from “the agent in my terminal” to runs they can start, check, cost and trust.
- Who it is not for
- People new to coding agents, and teams who want a product or a hosted service rather than the mechanics.
- Format
- A live online cohort of up to 12, one lab per session, eight sessions over four weeks.
- Where the work happens
- A browser workspace (Educates) per participant on one reference repository (“Ticketdesk”, Python 3.13 and Django 5.2).
- You bring
- A browser, an API key from the Claude Console (about $1.50 of model calls for the scripted steps, plus your interactive work), Python, git and shell.
- Language
- The labs are written in German.
- You walk out with
- A factory on your own repository: reproducible build, work packets, sensors and hooks, a sandbox, holdout verification, an orchestrator, metrics and budgets, and a review policy that decides what merges without a human.
- Price
- Not fixed yet.
Labs#
All eight labs are written and were walked through end to end on 2026-10-03; none has run for participants yet.
- The factory model and a reproducible repository (75 min)
- The work packet (90 min)
- The harness: sensors and hooks (90 min)
- Sandboxes and security (90 min)
- Verification (85 min)
- Orchestration (110 min)
- Operations: metrics and budgets (90 min)
- Autonomy and the capstone (120 min): five packets go through the factory, and one merges without review by your own policy