Most AI agent failures are really data-quality failures. Here is a practical guide to cleaning inputs, structuring knowledge, and designing workflows so agents can make useful decisions without creating expensive messes.