Making AI Agents Safe and Fast- Jev, Obsidian, and the Meta-Harness

Japanese Edition
https://daisuke.masuda.tokyo/article-2026-09-22-1843

This deck explains how to operate AI agents as a governed team by separating knowledge, decision-making, and execution. Obsidian serves as the source of truth for durable context, while Jev provides fast, typed probabilistic decisions for routing, risk scoring, and approval gates. Claude Code performs implementation work and connects through MCP to tools such as GitHub, Jira, CI/CD platforms, and internal wikis. Practical scenarios include incident triage, safer releases, and documentation synchronization, governed through explicit ALLOW, ASK, and DENY policies.

Transcript