Mihail Eric Maps Software Factory Components for Stanford CS146S Course

Mihail Eric, instructor of Stanford's CS146S: The Modern Software Developer course, presented a software factory component diagram in which agents independently launch repositories and every change passes the same gate of CI, static analysis, tests and AI review. The diagram defines factory rules in code, runs an orchestration loop from triage through spec, implementation, review and verification to pull request, merge, deployment and monitoring, and shifts human roles from executors to risk-based managers.
Event version 1.
Key insights
Factory definition managed in code
The top layer of the diagram is the factory definition, managed in code through skills, rules and AGENTS.md, and automation configuration that constrain every stage.
Orchestration loop and release pipeline
The middle orchestration loop runs triage, spec, implementation, review and verification, then moves to pull request, merge, deployment and monitoring of errors, alerts and usage. Specs require human approval, reviews are risk-tiered, and implementation runs in isolated sandboxes without production credentials.
Single shared gate for all changes
Every change passes the same gate of CI, static analysis, tests and AI review, with an AI judge scoring the full run against the spec. Failures are written back into the system as documentation, skills, check rules or tools to lower defect rates over time.
Human role shifts to management
Humans move from executors to managers: planning synchronously, executing asynchronously, and routing reviews by risk, while security, data and payments remain under human control.
Metrics focus on cost and intervention
The bottom metrics are cost per pull request, cycle time and number of human interventions; lines of code and tokens are not among the listed metrics.