When and how often to run evals

Your AI coding agent learned from millions of lines of code, including insecure ones. That means by default, it can write vulnerable code too.

So how do you fix that?

John Groetzinger, Principal Engineer at Cisco, built CodeGuard, a security skills layer that teaches coding agents how to write and review code securely. He tested it against real scenarios.

The result:
84% success rate vs 47% baseline. Nearly 2× improvement.

In this episode we get into:

• how CodeGuard works
• why Cisco open sourced it
• the surprisingly simple method that gets agents to fix their own mistakes

Try CodeGuard: cisco/software-security on the Tessl registry.

Connect with us here:
John Groetzinger: https://www.linkedin.com/in/john-groetzinger/
Cisco: https://www.linkedin.com/company/cisco/
Simon Maple: https://www.linkedin.com/in/simonmaple/
Tessl: https://www.linkedin.com/company/tesslio/

Join the AI Native Dev Community on Discord: https://tessl.co/4ghikjh

Ask us questions: podcast@tessl.io