I built web analytics with no dashboard, only an MCP
Building a unified control plane for operational intelligence is challenging due to LLM hallucinations, the need for a structured semantic layer over raw data, maintaining context purity across domains, and ensuring universal connectivity. These issues require architectural commitments like treating AgentOps as a first-class discipline and developing a living semantic layer rather than a static catalog. An alternative approach to traditional dashboards involves using AI coding agents that directly query tools for analytics, providing context for tasks like code development or deployment monitoring without requiring manual data interpretation. AI
IMPACT Highlights key challenges in developing sophisticated AI agents and control planes, informing operators about the complexities of operationalizing AI.