Topic roadmap
Agents
15 lessons, about 47 hours in all, from the basics to advanced. Start with the first, or jump to what you need.
Lessons, easiest first
- Agents · Intermediate 3 · about 3 hAgent memoryGive an agent the right memory: conversation state for the current task, long-term facts and past outcomes stored for later, and rules for what to forget.
- Agents · Intermediate 3 · about 3 hDesigning tools for agentsDesign tools a model uses correctly: few and purposeful, clearly named and described, with tight parameters, useful results and errors the model can act on.
- Agents · Intermediate 3 · about 2 hHow agents failName the common ways agents fail and add guards against each one.
- Agents · Intermediate 3 · about 3 hHuman approval stepsDecide which agent actions need a person's approval, build the pause, the approval screen and the resume, and keep approvals from becoming a rubber stamp.
- Agents · Intermediate 3 · about 5 hLangGraph: agents as graphsBuild an agent in LangGraph with explicit state, nodes and edges, checkpoints that let it pause and resume, and conditional routing.
- Agents · Intermediate 3 · about 3 hLogs and traces for agentsRecord every model call and tool call an agent makes, so you can see why it did what it did.
- Agents · Intermediate 3 · about 4 hMCP servers and clientsExplain how the Model Context Protocol connects AI apps to tools and data, build a small MCP server, and know the security questions to ask.
- Agents · Intermediate 3 · about 3 hPlanning and reflectionChoose between ReAct, plan-then-execute and reflection for an agent task, and know what each costs in calls, latency and reliability.
- Agents · Intermediate 3 · about 4 hThe agent loopBuild a loop where a model plans, calls tools, reads results and repeats until the task is done.
- Agents · Pro 1 · about 3 hAgent SDKs and sub-agentsKnow what vendor agent SDKs provide, when to use one instead of your own loop or a graph framework, and when splitting work across sub-agents pays off.
- Agents · Pro 1 · about 3 hBrowser and computer-use agentsDecide when an agent should operate a browser or desktop like a person, build it safely with scripted tools first and screen control last, and contain the risks.
- Agents · Pro 1 · about 2 hLoop limits, budgets and kill switchesPut hard limits on agent steps, tokens, time and spend per task, user and day, and build a kill switch that stops an agent or a feature at once.
- Agents · Pro 1 · about 3 hMulti-agent systemsChoose and build the main multi-agent patterns (supervisor with workers, hand-offs, and reviewer agents) and know the failure modes each brings.
- Agents · Pro 2 · about 3 hAdvanced harness engineeringImprove an agent by changing what surrounds the model: the loop and stop rules, tools, context handling, files on disk, verification gates, permissions and budgets, and prove each change on a fixed task set.
- Agents · Pro 2 · about 3 hHarnesses for long-running agentsDesign a harness that keeps an agent working for hours: planner, generator and evaluator roles, a feature list and progress file, context resets, checkpoints that survive a crash, git as memory, budgets and clear points to stop and ask a human.
On these role roadmaps
Interview questions on agents
- What are common failure modes of agents, and how do you guard against them?
- What kinds of memory can an agent have, and how do you implement them?
- Your agent ran overnight and ran up a large API bill. What probably happened, and what do you change?
- How would you design a harness for an agent that has to work on one task for several hours?
- What is the difference between an agent and a workflow?
- What are computer-use or browser agents good for, and what are their risks?
- An agent completes a task correctly most of the time but sometimes goes badly wrong. How do you debug it?
- What makes a good tool for an agent?