Evaluate before deploying Monitor after deploying Use what you learn to make the next version better

This post emphasizes a three-stage lifecycle for AI/LLM applications: evaluating systems before deployment to ensure quality, monitoring their performance in production to catch issues, and using the insights gained to iteratively improve subsequent versions. The advice advocates for a continuous feedback loop that treats deployment as part of the development process rather than an endpoint.

Source: Harrison Chase (X) | 2026-05-09