@abdulhamidsonaike: Your AI system can return a 200 status code and still be failing. Most teams check AI the same way they check any other app. 200 means it's fine. 500 means something broke. But with AI, that isn't enough. The server can be up while the answers are wrong, too costly or unsafe. Here's what you should be watching instead: 1. Hallucinations Is the model making things up? Track how often answers don't match the facts or the source. 2. Cost per feature Every chatbot, summary or landing page costs tokens. Know what each feature costs you to run. 3. Token use per person See how many tokens each team member uses. It helps you spot waste and repeated work. 4. Security Watch for jailbreaks, prompt injection and personal data leaking in or out. 5. RAG quality If you feed your own documents to the AI, check that it finds the right ones and uses them correctly. A green status light tells you the system is running. It doesn't tell you the system is working. What do you track in your AI systems today? #AI #LLM #AIObservability #RAG #AIEngineering
