The jump from engineer (implements components) to architect (designs the end-to-end system and owns the tradeoffs). No new ML here — you learn to specify, select, scale, govern, and justify AI systems, and to produce the artifacts an architect is judged on: C4 diagrams, ADRs, decision matrices, cost models, and governance checklists. Do this after Phases 6–8 — you architect systems you've already built. Portfolio of decisions > certificates.
Pick one system from your projects/ (the support copilot proj-01 and the permissioned
agent sandbox proj-08 are the richest) and produce a portfolio-grade design package —
the exact artifacts an AI Architect is hired on: a C4 architecture diagram (components,
interfaces, state, failure points), the key ADR with options and consequences, the
decision matrix behind it, a monthly cost model at a stated volume with the break-even,
and a governance checklist (NIST AI RMF + EU AI Act tier + success metrics). Bundle them as
docs/architecture/ in that project's repo. Ready when you can take any AI problem
and, on a whiteboard, sketch the system, name the failure modes, state build-vs-buy with its tradeoffs,
estimate cost at volume, name the compliance tier — and explain it all in cost/risk/outcome terms.
Full guide: phase-9-architecture.md ↗ · templates: ADR / decision-matrix / cost-model / governance ↗