Addy Osmani has a phrase for the moment engineering is in: “cognitive debt.” It’s the gap between what an agent has generated and what any human on the team actually understands. And it’s one of the problems leaders have to solve before they can trust agents at scale.
In this conversation, AI Tech Talks host Morten Rand-Hendriksen sits down with Addy—an AI engineering leader and the creator of the open-source Agent Skills project—to work through what it takes. They start with the paradox at the heart of the shift: code generation got cheap, but quality didn’t. Quality moved into the constraints around your architecture, and someone still has to be able to explain what changed, why it was safe, and what happens if it fails.
From there, Addy walks through the vocabulary engineers are using to describe agentic work: what a harness actually is, how loops let one good run become a process you can trust to run again, and why the inner loop belongs to the agent while the outer loop stays with you. He also names the failure modes—cognitive surrender when engineers stop evaluating output, orchestration tax when they try to run too many agents at once—and makes the case that codified team skills carry standards forward better than any individual prompt.
This is a recording of a live event and features audience participation and an unscripted Q&A. Visit linkedin.com/learning/topics/linkedin-live for more conversations featuring prominent leaders.
Learn More