A syllabus for the agentic era, 206 links deep

What it does A curated learning roadmap — not a library — covering the agent loop, MCP, context and memory, orchestration, evals, and production practices. It bundles 206 resources from OpenAI, Anthropic, Google, and community sources into phases, each with Read First, Then Read, and Build Exercise checklists. Phase 0 is a hands-on lab that rebuilds a Claude Code-style agent from a bash loop up through planning, subagents, and skills, with runnable Python alongside the notes.

The interesting bit It’s structured for people at different starting points: newcomers follow phases linearly, practicing engineers jump in at MCP or evals, and project-builders go straight to the applied tracks (coding agents, security, code review, SRE). The “should I build an agent?” four-question checklist from Anthropic — complexity, value, capability, error cost — is a genuinely useful gate before you write any agent code at all.

Key highlights

• Phase 0 lab builds an agent loop incrementally: bash-only, model-as-agent, todo planning, subagents, skills
• Skills map ties the roadmap to Andrew Ng’s AI engineering framework
• Applied practice tracks for coding agents, security, code review, and SRE
• Priority-tagged reading table (P0/P1/P2) so you can triage what to actually read
• Includes mental models like “think from inside the agent’s ~10K–20K token context window”

Caveats

• It’s a reading list with exercises, not a framework or tool — the value is the curation and ordering, not code
• Phase 0 tutorials are adapted from shareAI-lab/mini-claude-code, so the hands-on portion is partly upstream work

Verdict Worth a bookmark if you’re transitioning from “I use Claude Code” to “I build agent systems” and want a path instead of a bookmark folder. Skip it if you already have a production agent in flight and just need a specific answer.