Stop giving your agents database credentials
A number came up at Data + AI Summit this week that validated something we've been saying at CrewAI for a long time: The agent loop, the reasoning, the tool calling, the clever prompting, is roughly 1% of the work. The other 99% is building, configuring, deploying, securing, evaluating, and monitoring. I was on a founders' panel at Summit called "The Agentic Stack," and that 99% was the actual conversation. Not which model is smartest, nor which framework wins but the actual infrastructure betwe