Phase 06 — Product and Operations
Deployment changes the system. Real traffic changes inputs and incentives; infrastructure adds latency and failure; interfaces shape user behavior; feedback loops change future data; and every model, prompt, dataset, tool, and policy needs a version and rollback path.
Mental model
Production is a feedback system: user decisions create traffic and data; infrastructure and models transform them; observations drive releases. Reliability requires traceable versions, bounded fallbacks, human ownership, and rehearsed recovery.
Roadmap through the branches
Exit criteria
You can frame the user decision, establish a non-AI baseline, budget latency and cost, trace end-to-end behavior, monitor drift and failures, route to fallbacks or humans, stage releases, and rehearse rollback and incident response.
Connects to: Phase 05 — Measurement and Trust · Labs, Research and Playbooks
Core sources
- Hidden Technical Debt in ML Systems — production coupling and feedback debt.
- Continuous Delivery for Machine Learning — reproducible releases and feedback.
- People + AI Guidebook — human-centered product controls.