While evals involve testing, their purpose isn't just to report bugs (information), like traditional QA. For an AI PM, evals are a core tool to actively shape and improve the product's behavior and performance (transformation) by iteratively refining prompts, models, and orchestration layers.

Building non-deterministic AI products fundamentally changes the PM role. Instead of creating detailed, rigid specifications, the PM's primary task becomes defining and codifying "what good looks like." This is done by repeatedly grading AI outputs to train evaluation systems and guide the model's behavior.

When using an LLM to evaluate another AI's output, instruct it to return a binary score (e.g., True/False, Pass/Fail) instead of a numbered scale. Binary outputs are easier to align with human preferences and map directly to the binary decisions (e.g., ship or fix) that product teams ultimately make.

To ensure product quality, Fixer pitted its AI against 10 of its own human executive assistants on the same tasks. They refused to launch features until the AI could consistently outperform the humans on accuracy, using their service business as a direct training and validation engine.

For complex cases like "friendly fraud," traditional ground truth labels are often missing. Stripe uses an LLM to act as a judge, evaluating the quality of AI-generated labels for suspicious payments. This creates a proxy for ground truth, enabling faster model iteration.

Do not blindly trust an LLM's evaluation scores. The biggest mistake is showing stakeholders metrics that don't match their perception of product quality. To build trust, first hand-label a sample of data with binary outcomes (good/bad), then compare the LLM judge's scores against these human labels to ensure agreement before deploying the eval.

A one-size-fits-all evaluation method is inefficient. Use simple code for deterministic checks like word count. Leverage an LLM-as-a-judge for subjective qualities like tone. Reserve costly human evaluation for ambiguous cases flagged by the LLM or for validating new features.

You don't need to create an automated "LLM as a judge" for every potential failure. Many issues discovered during error analysis can be fixed with a simple prompt adjustment. Reserve the effort of building robust, automated evals for the 4-7 most persistent and critical failure modes that prompt changes alone cannot solve.

The prompts for your "LLM as a judge" evals function as a new form of PRD. They explicitly define the desired behavior, edge cases, and quality standards for your AI agent. Unlike static PRDs, these are living documents, derived from real user data and are constantly, automatically testing if the product meets its requirements.