Public pages
Deploying an agent with a one-line instruction and hoping for the best is not a strategy. This talk looks at the shared responsibility model for AI agents on Gemini Enterprise Agent Platform: which guarantees the managed platform gives you (Agent Sandbox and Runtime, Agent Identity and IAM, Agent Gateway, its policy engine, and observability and traces), and which controls you still have to build into the agent yourself: custom tools with no shell access, least-privilege identity, a fixed call graph and tool-callback deny rules.
Using a diagnostic SRE agent for end-to-end test failures in an immutable, multi-environment platform as the running example, the talk covers how the agent fits into the engineer’s SDLC feedback loop, and how a verifier enforces data provenance and semantic integrity. It catches overreach, broken references, unverified claims and malformed reports before the agent’s conclusions reach an engineer.
Event: AAIF Sydney Meetup
Date: 3 November 2026
Location: Sydney