Building real projects teaches AI skills faster than following structured curricula

A structured 30-day AI mastery roadmap follows a progression from prompting through data to applications. But mapping a real builder’s journey against the same roadmap revealed they’d already mastered context engineering, AI-assisted coding, and RAG fundamentals — not through study, but through solving real deployment problems, writing specs, and debugging integration issues on a production system serving 700 people.

This is the experiential version of what Compound engineering makes each unit of work improve all future work describes systematically: each real problem solved compounds into intuition that no curriculum can replicate. The builder learned context engineering by isolating Railway debugging from other work, not by reading about the four strategies (write, select, compress, isolate). They learned Spec files are external memory that survives context resets by needing spec files to survive context resets during multi-session debugging, not by studying best practices. The implication for AI-native education: provide scaffolded real projects, not structured lessons. This is the experiential case for why Don't be the discriminator — be the patron, not the judge — building in friction with the tool (patron) teaches more than selecting from AI output (discriminator).

Vivek’s “how to be good at research” frames the long-run version: “knowledge and productivity compound like interest,” and “the daily edges look trivial in isolation… give them a few years and they produce careers that look like luck from the outside.” Two of those edges are corollaries worth their own nodes — A clear public explanation is a genuine contribution and an unfakeable credential (public writing compounds sideways into collaborations and roles) and Your first subfield is an accident of timing — wander across several before you settle, because breadth is insurance (paying tuition in several corners by building, before specializing). The unifying instruction is Hamming’s: keep an open door, because the interruptions carry information about what the world actually needs.