SSanathinaiqa.hashnode.dev·Sep 29 · 3 min readAdding a Tool to Your AI Agent? Here's What Actually Needs RetestingTwo changes account for most of the "wait, when did that break" moments teams have with AI agents: adding a tool or widening a permission, and updating the model, system prompt, or orchestration layer01M
SSanathinaiqa.hashnode.dev·Sep 25 · 4 min readAI Agents Are Confused Deputies: Designing and Testing the Authority LayerAn old security bug, a new kind of deputy, and how to test for it In 1988, Norm Hardy described a compiler that could be tricked into overwriting a billing file. The compiler had permission to write t12M
SSanathinaiqa.hashnode.dev·Sep 24 · 9 min readHow BotGauge Is Building the Missing Layer for Trustworthy AI AgentsIn March 2026, researchers at Palo Alto Networks' Unit 42 documented something that should worry anyone shipping an AI agent: a web page, crawled in the normal course of an agent's job, contained text00
SSanathinaiqa.hashnode.dev·Aug 22 · 7 min readYour AI Test Suite Is Healing Itself. That Is the Problem.TL;DR: Self-healing tests sound like a feature. In practice they create a category of bug that is harder to catch than a failing test: a passing test that is asserting the wrong thing. Here is why it 00
