Rollout, safety and operations for AI features
Shadow mode and canary
Shadow mode and canary releases
- A Regression Suite for Prompt Changes That Won't Break Refunds
- How to Build a Shadow Test for a New Model Before You Switch
- Setting the Gate That Widens an AI Feature to Everyone
- Shadow Mode for a Candidate Prompt: Running It Beside Live
- Turning a Go Decision into a Production Rollout Plan
- When to Widen an AI Feature to Users Who Never Opted In
- Widening an AI Feature from 5 Percent to Everyone
- A Canary Release Checklist for an Agent That Moves Money
- All of Shadow mode and canary
Rollback and kill switches
Rollback and kill switches
- Writing a Kill Switch for an AI Feature
- Writing a Shutdown Section for an Agent Runbook
- How to Write a Kill Switch for an AI Feature
- Revoking an Agent's Access Without Breaking Other Systems
- Rollback Triggers for AI Features: Numbers Before Launch
- Testing Whether the Switch Works When the Queue Is Full
- The Dependency Map for an AI Feature
- When a Default-On AI Feature Breaks: A Rollback Plan
- All of Rollback and kill switches
Stop rules and shutdowns
Stop rules and shutdowns for pilots
- When a Pilot Hits Its Stop Condition: A Shutdown Checklist
- Who Owns the Stop Decision? A RACI for AI Pilots
- A Delegation Chain for Stop Authority When the Owner Is Away
- A Top-Tier Stop Rule the Next On-Call Can Run
- Designing the Timeout Fallback When a Stop Fires Mid-Run
- From Stop Rule to Shutdown: Ending an AI Pilot
- Handing a Stop Decision to the Next Team
- The Tabletop Drill for Stopping an Agent Mid-Run
- All of Stop rules and shutdowns
Logging and incident response
Logging and incident response
- How to Write a Decision Record for an AI That Affects People
- What a Decision Record for AI Integration Needs to Survive Audits
- What to Log When an AI Agent Makes a Decision: Fields and Format
- A Gap Event Schema That Survives a Workflow Rebuild
- A Schema for the Decision Record: Fields and Review Hooks
- Capturing the Stop in an Incident Log Reviewers Can Replay
- Generating Canary Data Without Leaking Real Records
- How to Re-Run a Decision Record After a Model Swap
- All of Logging and incident response
On-call and severity
On-call and severity for AI services
- How to Test an Escalation Path with a Real Ticket
- A Postmortem Template for a Missed Response-Time Target
- A Service Level Template for Probabilistic Systems
- A Staging Page: Rehearsing an AI Escalation Without Waking Anyone
- Choosing the Percentile for a Response-Time SLI: P50, P95 or P99
- How to Staff an On-Call Rotation for an AI Agent Vendor Contact
- How to Write a Severity Matrix the Model Can Apply
- Reading an AI Vendor's Severity Definitions Before You Trust Them
- All of On-call and severity
Ownership and service catalogs
Ownership and internal AI service catalogs
- How to Audit an Internal AI Service Catalog
- How to Price Internal AI Automation as a Service to Other Teams
- What to Include in an Internal AI Service Catalog Entry
- Consuming an AI Service Without Reading Its Catalog
- Cost Units for AI Services: How to Define and Verify Them
- How to Assign an Owner and a Metric to Each AI Step on the Map
- How to Hand an AI Agent from a Pilot Team to Daily Operations
- How to Price an Internal AI Service to Other Teams
- All of Ownership and service catalogs