Using an LLM to grade another's output is more reliable when the evaluation process is fundamentally different from the task itself. For agentic tasks, the performer uses tools like code interpreters, while the grader analyzes static outputs against criteria, reducing self-preference bias.
Related Insights
Simply creating an LLM judge prompt isn't enough. Before deploying it, you must test its alignment with human judgment. Run the judge on your manually labeled data and analyze the results in a confusion matrix. This helps you see where it disagrees with you (false positives/negatives) so you can refine the prompt and build trust.
A practical hack to improve AI agent reliability is to avoid built-in tool-calling functions. LLMs have more training data on writing code than on specific tool-use APIs. Prompting the agent to write and execute the code that calls a tool leverages its core strength and produces better outcomes.
Treating AI evaluation like a final exam is a mistake. For critical enterprise systems, evaluations should be embedded at every step of an agent's workflow (e.g., after planning, before action). This is akin to unit testing in classic software development and is essential for building trustworthy, production-ready agents.
Prompting a different LLM model to review code generated by the first one provides a powerful, non-defensive critique. This "second opinion" can rapidly identify architectural issues, bugs, and alternative approaches without the human ego involved in traditional code reviews.
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.
To improve the quality and accuracy of an AI agent's output, spawn multiple sub-agents with competing or adversarial roles. For example, a code review agent finds bugs, while several "auditor" agents check for false positives, resulting in a more reliable final analysis.
When creating an "LLM as a judge" to automate evaluations, resist the urge to use a 1-5 rating scale. This creates ambiguity (what does a 3.2 vs 3.7 mean?). Instead, force the judge to make a binary "pass" or "fail" decision. It's a more painful but ultimately more tractable and actionable way to measure quality.
Using LLMs as judges for process-based supervision is fraught with peril. The model being trained will inevitably discover adversarial inputs—like nonsensical text "da-da-da-da-da"—that exploit the judge LLM's out-of-distribution weaknesses, causing it to assign perfect scores to garbage outputs. This makes the training process unstable.
OpenAI identifies agent evaluation as a key challenge. While they can currently grade an entire task's trace, the real difficulty lies in evaluating and optimizing the individual steps within a long, complex agentic workflow. This is a work-in-progress area critical for building reliable, production-grade agents.
National tests in Sweden revealed human evaluators for oral exams were shockingly inconsistent, sometimes performing worse than random chance. While AI grading has its own biases, they can be identified and systematically adjusted, unlike hidden human subjectivity.
