23 short videos taking you from your first ConversableAgent to a working multi-agent system. Open-source. Episode by episode.
or browse all 23 trails ↓Each episode is a short trail, start at the trailhead, finish with running code. Live trails never close. New ones blaze open as they're cut.
Understand what an agent is, and why teams of agents beat monolithic prompts.
Install AG2, configure a model, ship a working agent in 20 lines.
Turn any Python function into an agent capability with the @tool decorator.
The two-agent pattern, an executor and a planner pass control back and forth.
Round-robin orchestration across a team of three or more agents.
Beyond round-robin, let the manager pick the right next speaker for the task.
Routing primitives, escalation paths, conditions, and when to break out of the loop.
Compose agents into supervised pipelines and hierarchical org structures.
Decision framework: two-agent vs. group vs. hierarchy, when each is the right call.
End-to-end build: triage, knowledge lookup, and human handoff across three agents.
Web-aware research agent that plans, queries, and synthesizes citations.
Wire a vector store into your agent and ground every response in your docs.
Headless browsing primitives, agents that read the live web.
Wrap any agent in a streaming chat UI with tool traces and citations.
Plug into the Model Context Protocol, Slack, GitHub, Notion, anything.
See what your agents actually do, traces, costs, and replay.
Sandboxes, allow-lists, and secrets hygiene for agents that run code.
Eval harnesses, golden traces, regression testing for non-deterministic systems.
Token budgets, model routing, caching that actually moves the bill.
Notebook → deployed service. Packaging, scaling, and health checks.
Run agents in parallel, score outputs, pick the best, fault tolerance for LLM calls.
Make agents think harder, branching reasoning and search-driven planning.
The AG2 roadmap and where to take your skills from here.
