Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Agent Factory Course

This course is a structured path through the Agent Factory wiki and examples: from definitions and loops to frameworks, production concerns, and advanced interoperability and self-improvement. It is designed for builders who learn best by reading, exercising, and shipping small vertical slices.

Who this is for

You should be comfortable with basic LLM usage (chat APIs, tokens, temperature) and general programming (read code, run scripts, use git). Prior exposure to HTTP APIs and JSON helps for tool and protocol modules. You do not need to be a researcher; the emphasis is practical systems.

Learning path

Work modules in order within each block unless a module’s prerequisites note otherwise. Blocks 1–3 establish vocabulary and architecture; Blocks 4–5 map those ideas to real frameworks and operations; Block 6 extends to protocols, harness engineering, and a capstone.

Estimated time: about 11–13 hours of focused study for all 23 modules (excluding capstone implementation time, which varies by scope).

Course structure: six blocks

Block 1: Foundations (Modules 01–04)

What agents are, architectures, the agent loop, and system prompts. Establishes shared language for the rest of the course.

Block 2: Core Building Blocks (Modules 05–08)

Tools, memory and context, planning and reasoning, error handling and recovery. These modules mirror what most production agents spend engineering time on.

Block 3: Patterns and Anti-Patterns (Modules 09–12)

Design patterns, multi-agent patterns, explicit anti-patterns, and state management. Use this block to critique designs before you commit to code.

Block 4: Frameworks (Modules 13–16)

Framework selection, LangGraph, OpenAI Agents SDK, and Anthropic-oriented patterns. Compare trade-offs rather than treating any stack as universal.

Block 5: Production (Modules 17–20)

Evaluation and testing, safety and guardrails, observability and debugging, deployment and scaling. Treat this block as the minimum bar before user-facing or high-impact automation.

Block 6: Mastery (Modules 21–23)

Protocols and interoperability (MCP, A2A), self-improvement and harness engineering, and the production capstone. Completing Block 6 means you can situate agents in an ecosystem, not only a single repository.

Module listing

Module Title Duration
01 What Are Agents 30 min
02 Agent Architectures 40 min
03 The Agent Loop 40 min
04 System Prompts for Agents 30 min
05 Tool Design and Integration 45 min
06 Memory and Context Engineering 45 min
07 Planning and Reasoning 40 min
08 Error Handling and Recovery 30 min
09 Agent Design Patterns 45 min
10 Multi-Agent Patterns 45 min
11 Anti-Patterns 35 min
12 State Management 35 min
13 Framework Selection 30 min
14 Building with LangGraph 45 min
15 Building with OpenAI Agents SDK 45 min
16 Building with Anthropic 45 min
17 Agent Evaluation and Testing 40 min
18 Safety and Guardrails 40 min
19 Observability and Debugging 35 min
20 Deployment and Scaling 35 min
21 Protocols and Interoperability 40 min
22 Self-Improvement and Harness Engineering 45 min
23 Capstone: Build a Production Agent 60 min

Using the wiki alongside the course

Each module points to concept and research articles under ../wiki/. Use the wiki when you need depth, citations, or alternative phrasing of the same idea. The ../wiki/examples/ tree contains good and bad exemplars; read bad examples to sharpen design reviews.

Start from ../wiki/INDEX.md if you prefer browsing by topic instead of linear modules.

Quality standard

Treat AGENT_SPEC.md as the repository’s cross-cutting quality bar: clarity of purpose, tooling discipline, safety, observability, and operational readiness. Use it when reviewing your own capstone or a teammate’s agent PR.

Autoresearch integration

Twelve modules (03, 05, 06, 08, 11, 12, 17, 18, 19, 20, 22, 23) include "Empirical note" sections drawn from a Karpathy-style autoresearch loop across 20 agents (~100 iterations). These sections connect each module’s topic to ranked, quantified findings about what actually raises agent quality scores. For the full ranked breakdown, see the Factory Showcase LEARNINGS.md or the distillate in the agent-factory README.

Companion project: Skill Factory

Agents and skills (procedures loaded by hosts) share engineering concerns: scoping, discovery, verification, and safety. The companion Skill Factory project lives at ../../skill-factory/ relative to this folder (sibling of agent-factory under the same parent workspace). Use it when you want parallel depth on authoring reusable SKILL.md assets and validation loops.


Agent Factory course README — aligns with modules 01–23 and the project wiki.