Evaluating AI Across the Research Pipeline

Our 2026 UXR Tools Map showed that most tools are now AI-native or augmented. But AI research tools don't make one decision — they chain many together, each step inheriting what the last one got wrong.

Research leaders Lindsey DeWitt Prat and Kaleb Loosbrock team up to tackle how we evaluate AI tools across the research pipeline. Lindsey introduces the “research risk cascade” and walks us through what she found tracking every divergence across a real pipeline. Kaleb joins for discussion on context engineering and what practitioners can start doing now.

You’ll learn:

• How errors compound across a research pipeline, and why “90% accurate” doesn’t mean what you think
• A practical framework for evaluating AI tools against what your research needs to preserve
• How context engineering can increase confidence in outcomes