2026 (rolling) — Anthropic Academy
Why this matters
anthropic.com/learn is Anthropic’s own structured-learning platform — branded Anthropic Academy internally — covering courses, guides, and certificates across the three “Claude paths” the company is targeting:
| Path | Audience |
|---|---|
| Build with Claude | Developers building Claude-powered applications |
| Claude for work | Organisations implementing Claude across teams |
| Claude for personal | Individual users |
Featured courses listed at the time of writing: Claude 101, Claude Code in action. Guides cover API development, the Model Context Protocol (MCP), and Claude Code specifically. Certificates are awarded on completion.
For a notes set focused on agentic engineering, Anthropic Academy matters as the first-party canonical pedagogy — what Anthropic itself says you should know, in the order they say you should learn it. Worth knowing about for two reasons.
Why first-party pedagogy is load-bearing
The 2026 ecosystem has a lot of community-authored learning material (third-party tutorials, Medium posts, YouTube channels, the writeups cited across this repo). Quality varies; accuracy varies more. The hooks-and-effort entry call-out of “13 plausible-sounding hook event names that don’t actually exist” is the canonical example of what happens when LLM-generated tutorials don’t get verified against first-party docs.
Anthropic Academy is the first-party reset point. When community material drifts (event names, API shapes, version-specific behaviour), the Academy is the version Anthropic stands behind. For a serious learner, the workflow is:
- Read the Academy material for the canonical first pass on any topic.
- Cross-reference the changelog for what changed since the Academy course was written.
- Then read community material for tacit knowledge, edge cases, and the “what does this look like in production” colour the official material rarely provides.
Doing it in the other order — start with community, only fact-check later — is how community-error patterns propagate. The 13-name “events that don’t exist” callout in the hooks entry is the case study.
Where it fits in this notes set
Reading this entry in particular, the implicit Academy-mapping across the repo:
| Academy course / guide | This repo’s parallel entries |
|---|---|
| Claude 101 | (not covered — this repo skips from-zero intros) |
| Claude Code in action | CLAUDE.md + plan mode, hooks, skills, subagents/managed-agents |
| MCP guide | skills + marketplace (where MCP fits in the plugin layout) |
| API guides | Out of scope; this repo focuses on the CLI / harness / agent layers, not raw API |
The Academy is broader (covers raw API usage); this notes set is narrower (focuses on the agent/harness/CLI surfaces). They overlap on the Claude Code material but don’t replace each other.
What’s notably not on Anthropic Academy
A useful framing: the Academy covers what Anthropic officially supports. What it doesn’t cover is where the action is for an agentic-engineering practitioner:
- Cross-vendor patterns. The Academy is Anthropic-shaped; agentskills.io, the ACP protocol (see the Hermes Agent entry), and local- inference backends (see the ds4.c entry) aren’t in the Academy curriculum even though they’re load-bearing in 2026 practice.
- Internal-Anthropic patterns from the leak. The source-leak entry surfaces architectural details that aren’t in any Academy course because Anthropic doesn’t (and probably shouldn’t) build curriculum around leaked internals.
- Economics-of-agentic-coding framing. Tom Blomfield’s observations (see Blomfield entry) aren’t pedagogy Anthropic would publish themselves — it’s external commentary on what their tools enable.
- Harness engineering as an organising concept. The Academy teaches the pieces (hooks, skills, agents) but not the umbrella — that crystallised in the harness engineering entry via community + arxiv work, not Anthropic curriculum.
This isn’t a complaint about the Academy — first-party pedagogy appropriately stays close to first-party surfaces. It’s a scope-completeness check: the Academy is necessary but not sufficient for agentic-engineering literacy in 2026.
Try this
If you’re new enough to Claude Code that any of the entries in this repo confused you on first read:
- Spend an hour on Claude 101 and Claude Code in action on the Academy. They’ll close the gap on prerequisites this repo assumes.
- Come back and re-read whatever entry confused you. It should land better with the Academy’s vocabulary in place.
If you’re already running Claude Code daily and find the Academy courses “below your level”:
- Do them anyway, on 2x speed if necessary. What you’re looking for isn’t the new-to-you material; it’s the gaps between what Anthropic teaches and what you’ve picked up from community sources. Those gaps are where the community- error patterns hide.
- Pair the Academy courses with the changelog page open. The courses are static; the platform moves. Reconciling the two is the actual learning.
Sources
- Anthropic Academy — the canonical landing page; structure may evolve over time
- Anthropic Claude Code changelog — the changelog is the always-current companion to the Academy’s snapshot-style course material
- Cross-references: every entry in this repo is in some sense an extension of what the Academy covers — community-authored notes on community-cited patterns. The Academy is the baseline; the rest of this repo is the delta