Resources

A working library for AI agent behavior.

Start with the framework, then choose the guide, checklist, example, or field note that matches the review in front of you.

Start here

The core framework for understanding what an AI agent becomes as it changes.

AI agent evaluation

The behavior-first way to evaluate AI agents through semantics, metrics, behavior diffs, evidence, and unknowns.

Guides

Practical ways to evaluate changes before release and inspect evidence in production.

Practical guide

Learn how to find AI agent regressions across prompts, models, tools, retrieval, and orchestration with behavior diffs and trace evidence.

Practical guide

A practical guide to evaluating AI agents in production with traces, privacy-safe evidence, coverage, baselines, and advisory review.

Practical guide

Understand the difference between AI agent observability and evaluation, and where each falls short.

Practical

A release checklist and a worked example for turning behavior changes into review decisions.

Checklist

Use this practical AI agent release checklist to review intended behavior, evidence, regressions, coverage, unknowns, and release risk.

Example

A refund-agent scenario where the eval score stays flat while pressure cues change the decision rule.

Field notes

Research and observations about the failure modes that aggregate scores can hide.

Field note

Why flat aggregate scores can hide changed AI agent behavior, affected workflows, and evidence gaps.