How to Build Better AI Evals with Claude Code in 5 Steps | Shreya & Hamel

By Peter Yang

Categories: Product, Startup

Summary

Writing is the final boss for LLMs—even with perfect evals, AI can't get it right in one shot. The key is externalize your judgment through top-down evals (task-critical criteria) paired with bottom-up evals (data-driven patterns), using agents to support error analysis before building automated LLM judges.

Key Takeaways

  1. Externalize your taste and judgment before writing evals—start by manually reviewing real traces and conversations to identify failure modes, not by asking Claude to generate evals from scratch.
  2. Use two-part eval strategy: top-down evals (well-defined criteria like character length, structure adherence) where LLM judges excel, and bottom-up evals (data-driven patterns you discover) where humans must lead.
  3. LLM judges have become highly accurate at evaluating specific, well-defined failure modes—replacing subjective scoring systems. Example: 'Is this output too long?' works better than vague quality ratings.
  4. Evals are iterative and improve through exposure to more failure modes—no single eval pass catches everything; manually review outputs, collect feedback patterns, then externalize those patterns into automated rules.
  5. Use agents to automate error analysis in the background while you manually review data, increasing leverage in the critical first stage of understanding what your system is getting wrong.

Related topics

Transcript Excerpt

You want to externalize your taste and your judgment before you start writing evals. >> No matter how many stuff I add and how many evals I add, it cannot get it perfect in one shot. >> Do you have the bottom up evals? I think this is where my question mark comes up for me. Bottom up is like data driven. Top down is not about the data. Claude is very very bad at coming up with bottom up evals. That's all you. The interface that it comes up with is going to be so much better than me looking at my data in say Google spreadsheets or something. We joke that writing is the final boss for LLMs. To like have it right like you in a way that you are satisfied with is extremely difficult. Hey everyone, my guests today are Hambo and Shrea, the instructors behind the most popular AI eval course online…

More from Peter Yang