Daily Kickoff

About

Up to seven things a day for people building with AI. Fewer when there aren't seven.

A quiet day — some candidates were dropped before publishing. 2 today.

  1. AWS Open-Sources Dogwood, Extending Cedar to Govern Sequences of Agent Tool Calls

    AWS open-sourced Dogwood, a policy language that extends its Cedar framework so rules can read an agent's history of prior tool calls, not just a single request. It ships under Apache 2.0 and is supported today in AgentCore Policy, though AWS says the reference interpreter is not ready for production use.

    InfoQ AI/ML agentic coding

    #
  2. How Uber evaluates AI agents at production scale

    At Arize:Observe 2026, Uber's Aayush Agrawal described a voice agent booking a ride to San Francisco International Airport that misread a child's background comment about wanting pizza as a request, and began rerouting the trip toward a pizza restaurant. The incident is cited to show production evaluation needs automatic tracing, living datasets, and shared ownership.

    Arize Blog agentic coding

    #