The team is adding customization at the agent layer, letting you tune models to follow instructions better. Newer tooling makes debugging easier across different setups. More capabilities roll out later this month.
How do we make our agent harness more flexible?
Why don't you let us customize the model for the agent layer?
So in the past, I haven't wanted to worry about the Spider-Man, whose bug is it, when there's an issue with how the agent performs. But what we've been really motivated by is the latest models are both better at instruction following, so they make fewer mistakes, but also now we have the tools to make it easy for customers and us to troubleshoot with models of all shapes and sizes. And agent harness designs of all shapes and sizes. We are going big on extensibility. Customers will see more on that later this month, and that is gonna be a huge unlock for so many use cases.
