Open reference library / 523 lessons
Study the mechanics. Build the system.
Explore the full English lesson sequence from AI Engineering from Scratch, alongside our governed agent field course. The original curriculum spans mathematics, machine learning, language models, tools, agents, infrastructure, safety, and capstones.
Reference material by Rohit Ghumare and contributors, MIT licensed. Imported at revision bf7791e14076. Runnable code and outputs remain linked to the source repository.
Connect your agent to this foundation ↗ — public search and lesson retrieval, with a reusable onboarding guide.
54 lessons
14 / agent engineering The Agent Loop: Observe, Think, Act ~60 minutes ↗14 / agent engineering ReWOO and Plan-and-Execute: Decoupled Planning ~60 minutes ↗14 / agent engineering Reflexion: Verbal Reinforcement Learning ~60 minutes ↗14 / agent engineering Tree of Thoughts and LATS: Deliberate Search ~75 minutes ↗14 / agent engineering Self-Refine and CRITIC: Iterative Output Improvement ~60 minutes ↗14 / agent engineering Tool Use and Function Calling ~60 minutes ↗14 / agent engineering Agent Memory — Virtual Context and Memory Paging ~75 minutes ↗14 / agent engineering Memory Blocks and Sleep-Time Compute ~75 minutes ↗14 / agent engineering Hybrid Memory: Vector + Graph + KV ~75 minutes ↗14 / agent engineering Skill Libraries and Lifelong Learning (Voyager) ~75 minutes ↗14 / agent engineering Planning with HTN and Evolutionary Search ~75 minutes ↗14 / agent engineering Anthropic's Workflow Patterns: Simple Over Complex ~60 minutes ↗14 / agent engineering Stateful Graph Orchestration — Durable Execution and Checkpoints ~75 minutes ↗14 / agent engineering The Actor Model for Agents — Async Messages and Typed Runtimes ~75 minutes ↗14 / agent engineering Role-Based Agent Teams — Roles, Tasks, Processes ~75 minutes ↗14 / agent engineering OpenAI Agents SDK: Handoffs, Guardrails, Tracing ~75 minutes ↗14 / agent engineering The Harness as a Library — Subagents and Session Store ~75 minutes ↗14 / agent engineering Production Agent Runtimes — Fast Instantiation and Typed Workflows ~45 minutes ↗14 / agent engineering Benchmarks: SWE-bench, GAIA, AgentBench ~60 minutes ↗14 / agent engineering Benchmarks: WebArena and OSWorld ~60 minutes ↗14 / agent engineering Computer Use: Claude, OpenAI CUA, Gemini ~60 minutes ↗14 / agent engineering Voice Agents: Pipecat and LiveKit ~60 minutes ↗14 / agent engineering OpenTelemetry GenAI Semantic Conventions ~60 minutes ↗14 / agent engineering Agent Observability: Langfuse, Phoenix, Opik ~45 minutes ↗14 / agent engineering Multi-Agent Debate and Collaboration ~60 minutes ↗14 / agent engineering Failure Modes: Why Agents Break ~60 minutes ↗14 / agent engineering Prompt Injection and the PVE Defense ~75 minutes ↗14 / agent engineering Orchestration Patterns: Supervisor, Swarm, Hierarchical ~60 minutes ↗14 / agent engineering Production Runtimes: Queue, Event, Cron ~60 minutes ↗14 / agent engineering Eval-Driven Agent Development ~60 minutes ↗14 / agent engineering Agent Workbench Engineering: Why Capable Models Still Fail ~45 minutes ↗14 / agent engineering The Minimal Agent Workbench ~45 minutes ↗14 / agent engineering Agent Instructions as Executable Constraints ~50 minutes ↗14 / agent engineering Repo Memory and Durable State ~60 minutes ↗14 / agent engineering Initialization Scripts for Agents ~45 minutes ↗14 / agent engineering Scope Contracts and Task Boundaries ~50 minutes ↗14 / agent engineering Runtime Feedback Loops ~50 minutes ↗14 / agent engineering Verification Gates ~55 minutes ↗14 / agent engineering Reviewer Agent: Separate Builder from Marker ~55 minutes ↗14 / agent engineering Multi-Session Handoff ~50 minutes ↗14 / agent engineering The Workbench on a Real Repo ~60 minutes ↗14 / agent engineering Capstone: Ship a Reusable Agent Workbench Pack ~75 minutes ↗14 / agent engineering Frame the Task Before the Agent Writes Code ~60 minutes ↗14 / agent engineering Build an Evidence-Backed Execution Plan ~65 minutes ↗14 / agent engineering Delegate Agent Work with Isolation and Merge Contracts ~70 minutes ↗14 / agent engineering Turn Every Agent Correction into a System Improvement ~65 minutes ↗14 / agent engineering Define the Outcome Before You Choose the Output ~60 minutes ↗14 / agent engineering Discover the Workflow People Actually Perform ~70 minutes ↗14 / agent engineering Map Assumptions and Resolve the Riskiest One First ~65 minutes ↗14 / agent engineering Choose the Smallest Slice That Can Change the Decision ~65 minutes ↗14 / agent engineering Write Specifications That Preserve Judgment ~75 minutes ↗14 / agent engineering Design Success Metrics Before the Result Exists ~70 minutes ↗14 / agent engineering Choose Prototype, Pilot, or Production Deliberately ~70 minutes ↗14 / agent engineering Build a Feedback Ratchet with Ownership and Retirement ~75 minutes ↗