Stop guessing how to fix your agent

Transcript

[00:00] If you’re building agentic applications,
[00:01] you’re likely aware of the new problems
[00:03] you don’t encounter in traditional
[00:05] software, from models making up
[00:06] information to agents calling the wrong
[00:08] tools or skipping required steps. And if
[00:10] you can’t see where things are going
[00:12] wrong, you’re left guessing what to fix
[00:14] and left hoping that your fix has
[00:15] actually worked. [music] That’s why
[00:17] building evaluations is a critical skill
[00:19] in AI engineering. I’m Lior Voss, head
[00:21] of developer relations at Arize [music]
[00:23] AI, and I’m excited to help you learn
[00:25] these skills in evaluating and improving
[00:26] agentic applications, a new course
[00:28] [music] built in partnership with
[00:30] deeplearning.ai.
[00:32] Through hands-on labs, you’ll learn a
[00:34] systematic approach to evaluating your
[00:36] agent, starting with the failures you
[00:38] find in its traces [music]
[00:39] and turning them into test cases you can
[00:41] rerun after every change. You’ll build
[00:43] code evaluators and LLM judges to catch
[00:45] those [music] failures and check whether
[00:47] those judges are reliable. Then you’ll
[00:49] use those evaluations to decide what to
[00:51] improve and check whether your agent is
[00:53] actually getting better. Join me in
[00:55] evaluating and improving agentic
[00:56] applications, and [music] let’s take the
[00:58] guesswork out of improving your apps.