Evals Are the New PRD: How Expedia's AI Chief is Reshaping Product Development

Evals Are the New PRD: A Paradigm Shift in AI Product Development

At VB Transform 2026, Xavi Amatriain, Expedia Group's first chief AI and data officer, made a bold claim that's reshaping how companies think about building AI products: evals are the new PRD (Product Requirements Document). This simple statement carries profound implications for how organizations develop, test, and deploy AI tools in the real world.

For decades, PRDs have been the backbone of product development. These documents outline what a product should do, who it serves, and how it should behave. But in the AI era, Amatriain argues that evaluation frameworks—or "evals"—are becoming the more critical blueprint for success. Instead of writing lengthy requirement documents, teams are encoding product specifications directly into comprehensive evaluation tests.

What Does This Mean in Practice?

The shift from PRDs to evals represents a fundamental change in how product teams approach AI development. Rather than describing features in prose, teams create concrete test cases that define exactly what the AI should and shouldn't do. These evaluations become the source of truth for product behavior.

According to Amatriain's comments, this approach encompasses multiple evaluation types, including:

• Functional evals that verify core features work as intended
• Red teaming evals that identify vulnerabilities and edge cases
• Security requirement evals that embed safety constraints directly into testing
• Performance evals that measure quality against benchmarks

By consolidating these requirements into eval frameworks, teams embed security, reliability, and performance considerations from the start—rather than treating them as afterthoughts.

Why This Matters for AI Tool Users

For anyone using or evaluating AI tools, this shift has significant implications. It means AI products are being built with more rigorous, objective testing frameworks in place from day one. When a company uses evals as their PRD, they're committing to measurable standards rather than vague promises.

This approach should lead to:

• Better product quality – Tools tested against comprehensive evals are more likely to perform reliably
• Improved safety – Red teaming evals help catch harmful behaviors before deployment
• Greater transparency – Eval-driven development creates clearer metrics for how AI systems actually perform
• Faster iteration – Teams can validate changes quickly against test suites rather than lengthy review cycles

The Broader AI Landscape Impact

This perspective from a major technology company like Expedia signals a maturing AI industry. As AI adoption scales, the need for rigorous evaluation frameworks has become obvious. Companies can no longer rely on informal testing or subjective assessments. The competitive advantage goes to organizations that can systematically define, measure, and validate AI behavior.

This trend is particularly important as enterprise AI adoption accelerates. Companies deploying AI in customer-facing applications, data processing, or decision-making systems need confidence that their tools will perform consistently and securely. Eval-driven development provides that confidence.

The shift also reflects growing industry awareness around AI risks. By treating security and safety requirements as core to the eval framework rather than optional add-ons, companies embed responsible AI practices into their development DNA.

The Takeaway

The evolution from PRDs to evals marks a maturing of AI development practices. For AI tool users and teams building with AI, this means better-tested, more reliable products. For the broader AI industry, it represents a necessary evolution toward more rigorous, measurable approaches to AI product development. As Amatriain's perspective gains traction, expect more companies to adopt eval-driven development—ultimately raising standards across the entire AI landscape.