Not Every Bug Needs an Eval

13 Oct 2026 · 1 min
0:31

Hamel Husain describes an LLM that couldn't schedule appointments because it didn't know today's date. The fix was simple: add the date to the prompt. He argues that not every bug needs a full evaluation, especially for straightforward issues like math or dates. Sometimes the fastest solution is the right one.

More from Hamel

47:59
youtube.com

Can Claude Build Your Evals?

2d ago
0:46
youtube.com

How to Get Your Team to Care About Evals

3d ago
12:47
youtube.com

7 Most Common AI Eval Mistakes (From Teaching 5K+ Engineers & PMs)

3d ago
40:01
youtube.com

How To Test AI Agents With Simulations

4d ago
0:43
youtube.com

Why Your LLM Judge Should Say Pass or Fail, Not 1 to 5

6d ago
0:39
youtube.com

Don't Just Slap a Chatbot on Your Product

1w ago
19:56
youtube.com

Is Opus 5.5 Better At Writing?

1w ago
1:22
youtube.com

How to Do Error Analysis When You Can't Look at the Data

1w ago
13:24
youtube.com

I Ranked Every AI Eval Concept From Worst to Best

1w ago
0:38
youtube.com

Why We Tell People to Read 100 Traces

1w ago
1:07
youtube.com

Your AI Tests Are Supposed to Go Stale

1w ago
See all content