Knowledge for Bringing AI into Production Operations
Covers the design decisions required to bring AI into operations — approval and rollback, audit logs, model costs, and failover when an external API fails.
- AI運用Read-onlyWhat it takes to bring an AI-generated application to productionWe organize the remaining work between "the code works" and "it can run in production" into 14 items, with commands for detecting hardcoded keys and auditing dependency vulnerabilities.Last Verified 2026-08-13
- AI運用Read-onlyWhy AI automated execution needs approval and rollback, and how to design for itDesign elements for automated execution — snapshots, approval gates, dry-run separation, allowlists, idempotency, audit logs, staged rollout, and a kill switch — are summarized, along with where to draw the line on what can run without approval.Last Verified 2026-08-13
- AI運用Read-onlyHow to Reduce Model Costs with an AI Router, and Failover Design for External AI API OutagesThis article summarizes a design for reducing model costs through task-difficulty-based routing and quality gates, along with a failover design assuming multiple providers. It does not cover pricing or cost-reduction figures.Last Verified 2026-08-13
- GIIP AI HarnessRead-onlyWhy GIIP AI Harness Excludes Query HintsWhy GIIP AI Harness excludes FORCE INDEX and join hints from SQL generation and tuning, and the steps to safely evaluate indexes marked for deletion.Last Verified 2026-09-06
- AI運用（自動生成コンテンツ全般）LowA guard that detects contamination in AI-generated content, and an operating rule that never treats regeneration alone as "done"This explains an operating guard that mechanically detects known contamination patterns in AI-generated content and never calls a regeneration "done" until someone has directly checked the resulting body text.Last Verified 2026-08-16
- TypeSafe AI Jev (System One Decision Model)Read-onlyWhat is TypeSafe AI Jev? — Fast "System One" Decision Model Features & ArchitectureUnlike text-generating LLMs, Jev is an ultra-fast System One decision model that returns structured choices and scores in a single pass. Designed by TypeSafe AI, it delivers 40-200x faster execution and 400x cost savings for agent routing and guardrails.Last Verified 2026-09-21