Infrastructure for Agents
Run a fleet of agents in production: compute, sandboxes, durable workflows, queues, state, tenant isolation, gateways and safe delivery.
2 modules · 8 lessons~16 hoursUpdated 2026-10-30
- Lessons8
- Quizzes8
- Sources49
- Of study~16h
About this course
About this course
An agent that works on your laptop still needs somewhere to run, a safe place to execute the code it writes, a way to survive crashes and week-long waits, storage for its state, credentials it can't leak, a route to several model providers, and a release process that doesn't break cases in flight. This course builds that platform for a fleet of per-step agents that review documents, draft replies, run identity checks and calculate payments, using the stacks AI builders actually pick today: Vercel Fluid compute, Sandbox, Workflow and Queues, Claude Managed Agents, Modal, E2B, Temporal, Inngest, Restate, Neon, Supabase, Upstash and the main AI gateways. Every lesson has costed worked examples, short Python code, primary sources and exercises you run with Claude.
About the running example
The examples follow a fictional property-law (conveyancing) firm in England and Wales and the other firms on the same platform, served by a fleet of per-step agents (search review, enquiries, ID checks, completion statements). 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.
- 1Choose where each agent runs (Fluid functions, containers, Modal, Managed Agents, eve on Vercel) from its duration, CPU profile, statefulness and cost, and cost the choice with real rates
- 2Run model-written code in a sandbox with the right isolation level, egress policy, credential brokering and lifecycle
- 3Design durable workflows and queue consumers that survive crashes, deploys and week-long waits, and reason precisely about replay, retries and at-least-once delivery
- 4Decide where each kind of agent state lives (system of record, workflow log, transcripts, files, coordination) and size it
- 5Isolate tenants and secrets end to end: short-lived credentials, row-level security, per-tenant keys and budgets
- 6Put an AI gateway in front of the fleet and ship agent changes through preview environments, eval gates and gradual rollouts
2 modules, 8 lessons.
9h 50m of lessons
Module 1 · 4 lessons · 4h 55m
Compute and execution
Where agent code runs, where model-written code runs, and how long-running work survives failure and waiting.
Module 2 · 4 lessons · 4h 55m
State, access and delivery
Where agent state lives, how tenants and secrets stay separate, how model traffic is routed and governed, and how changes reach production safely.
A production platform slice for one agent
Build the infrastructure under one agent from the fleet (a payment calculator), end to end and on synthetic data. A turn handler, a sandboxed calculator, a durable workflow fed by an idempotent queue, tenant-isolated state, a gateway with per-customer keys and budgets, and a CI pipeline with an eval gate.
More in Production AI at Scale.
- AdvancedAI · Reliability
Scaling AI Systems
Keep a fleet of agents fast and available under real load: rate limits, flow control, retries, backpressure, caching, failover, latency budgets, load tests and capacity plans.
7 lessons ~18hFree - 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