Level 11 of 14 · about 5 weeks

Production architecture and system design

Run agents as services: queues, durable execution, streaming, cost control and the design reasoning interviewers test.

Modules

1. 1

   APIs for long-running agents

   Job IDs, status endpoints, cancellation and signed webhooks

   Core 3 parts · about 3 hours

2. 2

   Queues and background jobs

   Workers, leases, retries with jitter, dead letters and fair scheduling

   Core 3 parts · about 3 hours

3. 3

   Durable execution

   Journals, replay, idempotent effects and surviving a crash mid-run

   Core 3 parts · about 3 hours

4. 4

   Streaming to clients

   Server-sent events versus WebSockets, resumable streams, heartbeats and backpressure

   Common 3 parts · about 3 hours

5. 5

   Cost, latency and routing

   Cost per resolved task, caches, model routers and per-task budgets

   Core 3 parts · about 3 hours

6. 6

   Reliability patterns

   Timeouts, retry budgets, circuit breakers, fallbacks, rate limits and bulkheads

   Core 3 parts · about 3 hours

7. 7

   Multi-tenant isolation

   Keeping one customer's data, cache, compute and credentials away from another's

   Common 3 parts · about 3 hours

8. 8

   Deployment and CI/CD

   Containers, secrets, eval gates, canaries and rollbacks

   Core 3 parts · about 3 hours

9. 9

   Versioning prompts and models

   Release manifests, pinned snapshots, replay and reproducing yesterday's behaviour

   Common 3 parts · about 3 hours

10. 10

    System design practice

    Designing a support agent platform and a research agent, with the trade-offs

    Core 3 parts · about 4 hours

11. 11

    Incident response

    Alerts, kill switches, runbooks and blameless post-mortems for agent systems

    Common 3 parts · about 3 hours