// Encode / Learn
Learn
Learning Claude by building. Six real, varied projects — one at a time, in public — each one earning the next concept as the build needs it. No typecasting: the lane gets picked at the end, from evidence, not guessed up front.
- P1 · Claude Platform M1.1–M1.3 — platform, messages.create, model choice Shipped
Single API call, wired into a real workflow
Session-notes → blog-draft tool. Built, shipped, live-tested, then pulled back to a local-only script.
- P2 · M2.1 agent loop · M2.2 tool use Now building
The agent starts taking actions
A small lookup-and-answer agent.
- P3 · M2.3 thinking · M3.1 built-in tools · M3.4 context management Planned
Thinking + keeping context straight
A different vertical than P2, on purpose — sample widely.
- P4 · M3.2 skills · M3.3 MCP Planned
Real data + a reusable skill
Connect to a real system worth caring about.
- P5 · M4.1 managed agents · M4.2 build your first one Planned
It runs itself
Promote whichever earlier project felt most alive.
- P6 · M5.1 building with Claude Code Planned
Capstone — built with Claude Code, and the lane decision
Review the signal from P1–P5. Pick the lane from evidence.
Every project ships a public case study before the next one opens. Signal — what came easy, what got traction — gets logged and feeds the lane decision after P5.