Ship it to real users. Keep it standing.
You are a builder. Level 3 makes you a systems designer: someone who ships AI to real users with service levels met, handles incidents and drift, survives adversarial inputs, and holds cost at scale. Four to six months of shipping, not reading.
What you can do when you finish
- You have shipped an AI product or feature to real users, with service levels met
- You have orchestrated a multi-agent workflow in production
- You have handled a real drift event or incident, or run a realistic dry-run of one
- You have run a model bake-off and made the call on measured results
- You have hardened one workflow against adversarial input
- For companies: full observability, incident response tested, cross-department scaling underway
What is inside
Fourteen modules on the Individual track covering production systems design, and ten on the Company track covering AI operations at organizational scale, plus a production checklist in the shared folder. Every module ends with the same question: did you ship the thing? Reading about incidents does not count as handling one.
Where this meets the doctrine: the guide's Order 6, Improve, asks one question of a one-person foundation: is it actually working? Level 3 is the industrial version of the same question, asked of systems other people depend on: observability, service levels, incident playbooks written before you need them.
How to start
- Pick one Level 2 project and take it to production.
- Add observability and service levels before customers touch it.
- Run a bake-off before you commit to a model.
- Red-team it, and write the incident playbook before the incident.
Read Level 3 on this site Start Level 3 on GitHub The whole book (PDF)