AI Systems in Production
Six weeks on operating AI systems after launch: tracing, cost control, model routing, incident response, and the deployment patterns that make failure survivable rather than public.
Join the waitlistWho should take this program?
You'll get the most from this if
- You have built an LLM or agent system
- Comfortable with containers and CI
This is not for
- Anyone who has not yet built something to operate
What you will build
- A deployed system with distributed tracing and cost dashboards
- A failure taxonomy built from your own production traces
- An incident runbook with rollback criteria
What you learn, week by week
Module 1 — Observability
- Instrument the full loop with spans
- Track tokens, latency and cost per request
- Decide what to alert on
Module 2 — Failure taxonomy
- Categorise real production failures
- Find the signals that predict them
- Prioritise by user impact
Module 3 — Cost control
- Profile spend honestly
- Cache and route between model tiers
- Set hard budget limits
Module 4 — Deployment patterns
- Run in shadow mode against live traffic
- Ship behind a canary
- Define rollback triggers in advance
Module 5 — Regression control
- Run evals in CI
- Version prompts like code
- Prevent silent behaviour drift
Module 6 — Incident response
- Write the runbook before the incident
- Practise a rollback
- Produce a postmortem that changes something
How this program handles evaluation and governance
This course is largely responsible engineering by another name. Shadow deployment makes failure free. Rollback criteria defined in advance stop judgement calls under pressure. A failure taxonomy turns anecdotes into evidence. And a postmortem that changes something is the difference between an organisation that learns and one that repeats.
Tools and stack you will use
Fees
Request pricing
Fees depend on cohort, format, and whether this runs for an individual or a team. We'll send the full breakdown.
Get fee detailsQuestions people ask
Is this an SRE course?
It borrows from SRE, applied to systems whose output is non-deterministic.
Do I need my own production system?
You need something you have built. We can supply a system to operate if yours is not ready.