Production Agent Systems
Take an agent from working demo to a system you can run for months: architecture, security, compliance, observability and safe change.
2 modules · 6 lessons~18 hoursUpdated 2026-10-30
- Lessons6
- Quizzes6
- Sources40
- Of study~18h
About this course
About this course
A demo agent answers one question; a production agent runs for the weeks a real case lasts, reads documents written by strangers, touches customers' money and personal data, and keeps working through model upgrades. This course covers the engineering that makes that possible, always landed in Claude Code, the Claude Agent SDK and eve: when multi-agent designs pay for themselves, durable sessions and idempotent side effects, prompt-injection defence by design, audit trails for regulated work, tracing and cost budgets, and the improvement loop that lets you upgrade a model without fear. Each lesson has worked numbers, primary sources and hands-on exercises on your Claude subscription, with eve agents on OpenRouter or a local model.
About the running example
The examples follow a fictional property-law (conveyancing) firm in England and Wales, whose agents run for the weeks a property transaction lasts, under the firm's regulatory duties. It's a case study, chosen because its documents are long, structured and high-stakes. You don't need a legal background, and the same patterns apply to contracts, insurance claims, invoices, support tickets or patient records. In the lessons, a matter is one client's case and a solicitor is the lawyer who signs work off.
What you'll learn.
- 1Decide when a multi-agent design beats a single agent, and estimate its token cost, latency and failure rate before building it
- 2Design durable, resumable agent workflows that wait on external events for weeks and perform each side effect exactly once in effect
- 3Threat-model an agent with the lethal trifecta and apply architectural prompt-injection defences: least privilege, sandboxing, approval gates and data isolation
- 4Build an audit trail and human sign-off flow that supports regulatory, anti-money-laundering and data-protection obligations, without overstating what the rules require
- 5Instrument agents with OpenTelemetry and eve Agent Runs, and set token, latency and cost budgets per case
- 6Run a continuous-improvement loop: online signals, drift detection, eval-gated model upgrades and canary releases
2 modules, 6 lessons.
7h 45m of lessons
Module 1 · 3 lessons · 3h 55m
Architecture
How the system is shaped: when to split work across agents, how to survive weeks of waiting and crashes, and how to stop untrusted documents steering the agent.
Module 2 · 3 lessons · 3h 50m
Operating agents
How the system is run: auditability for regulated work, traces and cost budgets, and changing models and prompts safely over time.
Production-readiness review and hardening
Take your drafting agent from the Building Agents course (or your own agent) through a production-readiness review, then make one concrete, tested fix in each area of this course, from injection defence to an eval-gated model upgrade.
More in Production AI at Scale.
- AdvancedAI agents · Infrastructure
Infrastructure for Agents
Run a fleet of agents in production: compute, sandboxes, durable workflows, queues, state, tenant isolation, gateways and safe delivery.
8 lessons ~16hFree - AdvancedAI · Cost engineering
LLM Cost Engineering
Know what each case costs in tokens, then cut that cost with caching, trimming, batching, routing, distillation and self-hosting, without losing quality.
8 lessons ~20hFree