Human in the Loop: Governing API Evolution in the Age of Agents

AI agents are now proposing, consuming, and evolving GraphQL schemas faster than any human review process was designed for. Subgraphs drift in days, ownership blurs across teams, and breaking changes ship into the supergraph before anyone notices.

This session is about a pattern for keeping humans in the loop without slowing agents down: proposal-based schema governance. Four primitives hold it together, which are ownership, change proposal, impact analysis, and approval. Every schema change, whether from a developer or an agent, flows through these before it touches the supergraph.

Through a live demonstration, you'll see this end-to-end. An agent proposes a schema change, the system routes it to the right subgraph owners, reviewers see the blast radius across consumers, and the change ships safely or gets rejected with reasoning the agent can learn from. Agents move fast, but humans still own the schema..