We are surrounded by AI that can talk, but not nearly enough that can actually do. In this episode of patchperfect, I sit down with my former business school classmate Glendon Hass, now a director at Bronson AI⁠, to unpack why we may still be in the stone age of AI and what it really takes for models to move from text boxes to real world tools.

We get into agentic AI, what practical tool use actually means, and why deep research is quickly becoming the new version of search. Glendon also explains why reinforcement learning can sometimes make models too agreeable, and why discernment may be the real human edge in an age increasingly shaped by AI.

This is the patch:

• what agentic AI looks like beyond chat interfaces
• why tool use is the real unlock for useful AI systems
• how reinforcement learning can shape model behavior in unexpected ways
• why human judgment still matters even as models get more capable

At its core, this episode is about shifting from AI that sounds smart to AI that can actually help us make better decisions.

We close with a lifestyle segment on Glendon's favorite reset ritual as both a director and a dad, which is baking cookies.

If you are curious about where AI is headed after the chatbot era, this episode will give you a grounded and practical way to think about what comes next.

All views expressed are my own.