Membership course · Intermediate
Autonomous AI Agents
Test narrow agent tools and local draft workflows, and practise decisions about permissions, recovery, budgets and scheduling.
Comfort reading basic code and using a terminal. Check the course-specific tool access and setup below.
Tum saath le jaoge
A tested local restock-draft boundary and a review of its limits. Live agents, durable recovery and deployment are separate follow-on work.
Specify agent goals, state, typed tools, evaluators, and hard stop conditions
Compare simple code, single-agent, pipeline, and multi-agent designs with evidence
Review MCP connection proposals for identity, tool scope and cross-tenant access
Work through supplied retry, checkpoint, approval, budget and scheduling cases
Run nine local draft-workflow assertions and document the untested production controls
Syllabus
15 lessons, in orderModule 1
Agent Fundamentals
1.1 From Chat to Agent: The 2026 Definition That Matters20m →1.2 Tool Use and Function Calling Explained Simply40m →1.3 Use Claude Code for a Bounded, Tested Repair25m →Module 2
Agent Frameworks
2.1 Comparing Agent Frameworks: When to Build vs. Buy25m →2.2 Model Context Protocol (MCP): Connecting Agents to Real Tools25m →2.3 Memory and State: Giving Agents Context That Persists25m →Module 3
Swarm Orchestration
3.1 Single-Agent vs. Multi-Agent: When Complexity Pays Off25m →3.2 Designing a Manager-Worker Agent Pattern25m →3.3 Handling Failures and Retries Across an Agent Swarm40m →Module 4
Operational Guardrails
4.1 Permission Boundaries: What an Agent Should Never Do Alone25m →4.2 Logging and Observability for Autonomous Workflows20m →4.3 Cost Controls: Preventing Runaway Agent Loops20m →Module 5
Prepare an Agent Workflow for Review
5.1 Plan a Scheduled Agent Run and Its Recovery25m →5.2 Human-in-the-Loop Checkpoints for High-Stakes Actions20m →5.3 Capstone: Test a Local Draft Workflow and Review Its Limits45m →Completion certificate
Mark every lesson complete and save your progress to your account. Generate a certificate with a unique verification ID. It records learner-reported completion, not a graded exam or reviewed project.
Your certificates →Tools & preparation
Claude Code · MCP · JavaScript
Comfort reading basic code and using a terminal. Check the course-specific tool access and setup below.
Live Claude Code activities require separate paid access: Claude Pro, Max, Team or Enterprise, or an Anthropic Console account with API billing. The free claude.ai plan does not include Claude Code. AI School membership does not pay for this tool access.
With course access but no paid Claude Code access, you can read the lessons, inspect supplied files, trace examples manually, and complete the specified local exercises and manual repairs. Label these results as manual preparation; they do not demonstrate a live Claude Code session or the full connected workflow.
Official Claude Code setup and account requirements →Third-party tools may have separate charges or limits. Check before paying.