Architecting Agentic Development Platforms

The internal developer platform isn't going away. But as agents move from assistants inside the IDE to autonomous actors triggering paths on their own, the architecture has to change. This session introduces the ADP reference architecture and shows exactly where and why.
Kaspar and Ajay break down the three layers of the ADP: the tooling layer (your existing IDP, now serving agents as a new consumer), the path specification layer (probabilistic, deterministic, and hybrid paths), and the agent infrastructure layer (the harness and governance plane that make agents safe to run at scale). They walk through eight canonical platform paths, from dispatch to remediation, and show how each evolves from human-driven to agent-executable.
To make it concrete, they trace a PR review through every infrastructure component: identity, dispatch, capability, context, execution, evaluation, security, and observability, showing which steps are deterministic, which are probabilistic, and why that distinction matters.
The session closes with a practical 90-day starting plan: two paths to build first, the minimum infrastructure required, and a key mindset shift from optimizing for speed to optimizing for throughput.