Zehrava Gate Blog
Writing about AI agent governance, the write-path problem, and building safe agentic systems.
Preventing Duplicate Tool Executions and Replay in Autonomous AI Agents
How to stop autonomous AI agents from executing duplicate actions, retrying side-effects, and creating race conditions using deterministic idempotency keys and execution gates.
Read →Semantic Guardrails vs. Deterministic Write-Path Gates: Why Prompt Filters Fail at the Action Layer
Understand why input/output prompt guardrails cannot guarantee action safety in autonomous AI agents, and how deterministic write-path control provides fail-closed action security.
Read →Human-in-the-Loop AI Agent Architecture: Async Approvals, Execution Orders, and State Recovery
A technical guide to implementing asynchronous human-in-the-loop approvals for autonomous AI agents without blocking worker threads or losing execution state.
Read →Why Autonomous Agent Governance Still Needs Human Approval
Microsoft and Rebuno shipped autonomous governance systems. Both include approval mechanisms but no dashboard. Here's why the 5% of agent actions that need human oversight require purpose-built tooling.
Read →Agent Governance: Zehrava vs. Microsoft vs. Alternatives
How Zehrava Gate compares to Microsoft Agent Governance Toolkit, Rebuno, NeMo Guardrails, Guardrails AI, LiteLLM, and Portkey. Complementary tools for a complete governance stack.
Read →What Is Write-Path Control for AI Agents?
AI agents that write, send, and execute without checkpoints create silent failure modes. Write-path control is the layer that puts a gate on those actions.
Read →How to Add Human Approval Gates to AI Agents
Step-by-step tutorial for adding human-in-the-loop approval checkpoints to AI agent write operations. Node.js and Python examples, 20–30 minutes to integrate.
Read →Why AI Agents Need an AI Agent Commit Checkpoint
An agent sent 47,000 emails in 90 seconds because no checkpoint existed between its decision and the production write — here's the pattern that prevents it.
Read →OpenAI Acquired Promptfoo. What That Means for Agent Governance.
Promptfoo tests what agents will do. Agent governance controls what agents are allowed to do. OpenAI's acquisition explains why the category is real — and what it does not yet solve.
Read →847 Records Overwritten, No Audit Trail: An AI Agent Production Incident
An enrichment agent overwrote 847 CRM records with partial data and left no trace — three incident patterns, and the structural fix that prevents all of them.
Read →Policy-as-YAML: AI Agent Policy Enforcement That Scales Across Teams
Define approval rules once in version-controlled YAML, share them across every agent and team, and give your compliance lead something they can actually read.
Read →