01
The Claude API
Messages, models, streaming, tokens and the caching decisions that cut your bill.
START FREE24 SLIDES · 45 MIN
AI Academy · AI-04 · Advanced
Managed Agents: Build production agents on the Claude API.
ACADEMY ACCOUNT
Free
with a free accountNo cardOne account unlocks this course and every other one, in full.
- This course8 modules · ~10h
- Pass unlocksEvery course
- FormatSelf-paced + hands-on
- CardNone needed
Build production agents on the Claude API: tool use, managed sandboxes, memory and guardrails.
Who it is for: Developers and technical founders building agent products or internal automation.
See it in action
Claude Agent SDK full workshop, by Anthropic
Syllabus
8 modules, ~10 hours, one capstone.
Module one is open to everyone, no account needed. The rest of the course opens the moment you create a free account, along with every other course.
02
Tool use and the loop
Tool schemas, the agentic loop and the SDK tool runner that runs it for you.
LOCKED20 SLIDES · 60 MIN
03
Managed Agents
Server-hosted agents with a managed sandbox: when they beat rolling your own loop.
LOCKED20 SLIDES · 60 MIN
04
Agent SDK deep dive
Build a custom agent product: sessions, subagents, permissions and hooks.
LOCKED19 SLIDES · 75 MIN
05
Memory and context
Long-running work: memory tools, compaction and context engineering.
LOCKED23 SLIDES · 60 MIN
06
Guardrails and evals
Permissions, human-in-the-loop gates and the eval harness that catches regressions.
LOCKED19 SLIDES · 60 MIN
07
Cost engineering
Caching strategy, model tiers, budgets and the metrics that keep an agent profitable.
LOCKED19 SLIDES · 45 MIN
08
Project + capstone
Build a production audit agent, then deploy your own with an eval harness.
LOCKED20 SLIDES · 120 MIN
CAPSTONE
Deploy an agent with at least three tools and an evaluation harness. Submit the endpoint.
What you walk away with
Skills that you can use straight away.
The Claude API fluently: Messages, models, tokens and prompt caching
Tool use and the agentic loop, hand-rolled and with the tool runner
Managed Agents running server-side in a managed sandbox
Memory, context management and guardrails for long-running work
A production agent deployed and handling real requests
TOOLS YOU WORK IN
Claude APIAnthropic SDKManaged AgentsClaude Agent SDKInside every lesson
Interactive learning activities.
Each lesson is a short deck of activity-book pages. You predict before you are told, do the task in your own store, and keep the notes. Built to be done, not skimmed.
Predict, then reveal
Commit a guess before the lesson gives you the answer, so it actually sticks.
Hands-on labs
Tick off real tasks in your own store, one at a time, with your progress saved.
Write-in workbook
Jot your answers straight into the page. They save on your device, yours to keep.
No walls of text
Flip-cards, compare panels and deep-dives instead of dense paragraphs to wade through.
Lab
LAB · MANAGED AGENTS
- Do the first task in your real workspace
- Tick each step off as you complete it
- Save the result as your deliverable
YOUR CALL FIRST
What will you try first?
Before you ask
Questions, answered.
Production agent engineering on the Claude API: the Messages API and prompt caching, tool use and the agentic loop, Managed Agents (server-hosted agents in a managed sandbox), the Agent SDK for custom agent products, memory and context management for long-running work, guardrails and evals, and cost engineering. The capstone deploys your own agent with an eval harness.
Anthropic's server-hosted agent runtime: your agent runs on Anthropic's infrastructure in a managed sandbox with tools, instead of you hosting and orchestrating the loop yourself. The course covers when they beat rolling your own loop, and when the Agent SDK or a hand-rolled loop is the better call.
This is a developer course. You should be able to write and run code in Python or TypeScript and be comfortable with APIs. It is the step up from our Claude Code course, which teaches building with AI; this one teaches building systems out of AI.
A production agent you designed, built and deployed: at least three tools, memory, guardrails, an eval harness that catches regressions, and unit economics you can defend. Plus the judgement to choose between Managed Agents, the Agent SDK and a custom loop per project.
All eight modules and every future update. API usage during the course is on your own key; the lessons are built to keep experiment costs in the tens of dollars, and the cost-engineering module pays for itself.
Start now. Lesson one, the Claude API deep dive, is readable with no signup. The rest opens with a free account, which covers every other course as well. No card, now or later.
AI-04 · FREE WITH AN ACCOUNT