Skip to content
LeetLLM
Tracks
Intermediate25 lessons

Agents and Tool Use

Build an assistant you could delegate useful work to, then learn how to inspect its progress and help it recover when an approach fails.

Engineers building coding agents, workflow agents, browser agents, or production tool-use systems.

You can design agent loops with tools, state, recovery policy, evals, and review gates.

  1. 1Prompt Engineering FundamentalsBuild an incident-triage prompt with clear roles, useful examples, structured outputs, evidence checks, and a separate authorization boundary.Preparation & PrerequisitesEasy16 min
  2. 2Calling LLM APIs in ProductionA notebook answer can be right and still stall, time out, or return HTTP 429. The server has to decide what to retry and what the user may see.Preparation & PrerequisitesEasy30 min
  3. 3Function Calling & Tool UseTrace a tool request from model to runtime, then build the checks that keep its actions, retries, and results under control.Applied LLM EngineeringMedium31 min
  4. 4Structured Output GenerationA build log needs a typed run id and status, but 'return JSON' can still come back as chat. Constrain the tokens, then check the facts yourself.Advanced Agents & RetrievalHard40 min
  5. 5MCP & Tool Protocol StandardsFollow a tool request from an AI host to an MCP server, then learn where compatibility, authorization, and trust checks belong.Applied LLM EngineeringMedium24 min
  6. 6ReAct & Plan-and-ExecuteA CI failure can change the next useful check. Choose one step at a time, or a plan you can revise when the evidence breaks.Advanced Agents & RetrievalHard64 min
  7. 7Context EngineeringAn agent can fit a 128k window and still repeat a disproven lock, so each request has to be curated and later work has to resume from checked artifacts.Applied LLM EngineeringMedium26 min
  8. 8Prompt Injection DefenseA report tells an agent to skip approval. Learn to quarantine its text, check actions in trusted code, and measure whether an attack caused harm.Applied LLM EngineeringMedium29 min
  9. 9Responsible AI GovernanceA blocked promotion is not yet governance. The release still has to name the workflow, its owner, the tested control, and evidence a reviewer can replay.Applied LLM EngineeringMedium32 min
  10. 10Guardrails & Safety FiltersOne assistant can answer a test question, pause an unapproved deploy, and stop a secrets request before it reaches a vault.Advanced Agents & RetrievalHard62 min
  11. 11Evaluating AI AgentsAn agent says it is done. Learn to check what changed, whether it respected permissions, and how reliably it can repeat the task.Applied LLM EngineeringMedium33 min
  12. 12LLM Observability & MonitoringSix responses return HTTP 200 and still include two unsupported approvals. Latency, cost, and traces matter only once that claim is visible.Applied LLM EngineeringMedium32 min
  13. 13Code Generation & SandboxingHelp coding agents turn runtime errors into tested repairs while keeping untrusted code away from host secrets and grading authority.Advanced Agents & RetrievalHard30 min
  14. 14Computer-Use / GUI / Browser AgentsAn agent reading a failed CI page can still click a deploy control, so the host must check each proposed action and verify the result.Advanced Agents & RetrievalHard51 min
  15. 15Human-in-the-Loop Agent ArchitectureAn agent proposes a risky change. Learn to pause it for review, preserve the exact proposal, and keep approval separate from the eventual write.Advanced Agents & RetrievalHard53 min
  16. 16AI Coding Workflow with AgentsA coding agent can return a green pull request that never fixes the webhook. Bound the change, replay the event, and keep the merge with a person.Advanced Agents & RetrievalHard35 min
  17. 17Agent Memory & PersistenceHelp a coding agent pick up yesterday's work using fresh, scoped notes and explicit authorization checks.Advanced Agents & RetrievalHard57 min
  18. 18Agent Failure & RecoveryWhen an agent times out, its action may still be running. Learn when to retry, verify state, slow down, or hand off.Advanced Agents & RetrievalHard33 min
  19. 19Multi-Agent OrchestrationUse multiple agents only when independent work earns their cost. Validate what they share, pause for review, and let one guarded writer act.Advanced Agents & RetrievalHard41 min
  20. 20Design an Automated Support AgentAlex asks for a $900 credit. Follow the invoice and policy checks that decide whether the agent can act or needs a specialist.Applied LLM EngineeringMedium39 min
  21. 21Model Versioning & DeploymentAn engineer reverts the prompt in Git, but the image and the evidence have already moved. The release ID has to name the combination that is actually answering.Applied LLM EngineeringMedium33 min
  22. 22Capstone: Document QAA published policy and a private note can both be retrieved. The lookup has to keep the note out and still cite only a passage that supports the question.Portfolio CapstonesHard54 min
  23. 23Capstone: Eval DashboardTwo document-QA runs can tie at 75% while one cites a private note. Build the dashboard that holds that run and keeps display filters out of the gate.Portfolio CapstonesHard38 min
  24. 24Capstone: Fine-Tuned ClassifierTrain an intake classifier that sends a freeze exception to a person, then hold the release when a required escalation is missed.Portfolio CapstonesHard47 min
  25. 25Capstone: Production AgentA policy question is not permission to restore a disabled key. Draft from bound evidence, and require a separate stored approval before any write.Advanced Agents & RetrievalHard33 min