How to Test AI Agents Before They Ship with Agent Assurance

How to test AI agents: point Agent Assurance at your agent's codebase. It finds the agent, writes the test scenarios, runs the agent for real, and grades every criterion against evidence, not against the agent's own account of what it did. 👉 Try Agent Assurance: https://www.testmuai.com/agent-assurance/?utm_source=youtube&utm_medium=organic&utm_term=VIDEO_ID&utm_campaign=agent_assurance_demo In this demo, Agent Assurance runs from the terminal as rook against a ticket-triage agent: ✅ /explore: finds the agent in the codebase and works out what it does ✅ /generate: writes features and scenarios, with criteria for each one ✅ /profile add: paste a curl command and rook learns how to call your agent ✅ /run: invokes the live agent and judges each criterion, with Pass, Fail or Unable to Verify ✅ The catch: the agent marked a PDF download bug as a P1 outage because "download" contains "down", and answered with full confidence ✅ Evidence for every verdict (expected, actual, quoted proof) in the local UI and the hosted web UI, with run history ⚡ Install: npm install -g @testmuai/rook (Node 22+) 📂 GitHub: https://github.com/LambdaTest/rook Agent Assurance is in pre-alpha: commands and file formats may still change. ⏱️ Chapters 00:00 Agent Assurance by TestMu AI 00:06 The agent under test: a ticket-triage agent 00:30 Install and sign in with rook login 00:42 Launch rook and create a project 01:00 /explore: rook finds the agent in your codebase 01:12 /generate: features, scenarios and criteria 02:18 Workflow status: what's done, what's next 02:36 Choose which scenarios to run 02:48 /profile add: tell rook how to call your agent 03:24 /run: test the live agent 03:36 Results: 2 scenarios fail, and why 04:16 Evidence for every criterion in the local UI 04:40 Run history in the hosted web UI 05:06 Status recap 05:12 Try it on your own agent #AgentAssurance #AIAgents #AITesting #AgenticAI #TestMuAI