Agent Infrastructure

Building on the three-layer ADP framework, this session takes a closer look at the agent infrastructure layer: the foundation that enables models to act, reason, and deliver useful output in a software development context.
Dilek breaks down the four elements of the agent harness: context, capabilities, execution, and eval, and explains how each shapes what an agent can do and the quality of its output. He then covers the governance pillars of identity, security, and observability, and how these extend across the full platform stack.
To ground the theory, he walks through a concrete example: a probabilistic PR review path, tracing each step through the ADP layers and showing how the infrastructure elements come together in practice.