From on-call to in-command

Speaking sessions

• Four primitives to evaluate and improve your on-call agent's autonomy

  On-call isn't one kind of work, so treating agent autonomy as a single switch breaks the problem. This session splits on-call into four primitives - triaging alerts, investigating them, acting on production, and learning from what happened - and shows why each one progresses at a different rate: investigation is bottlenecked on quality, action is limited by trust.

  We walk through how we build each primitive into the agents at Resolve AI, the shadow diff we use to tell which limit you are actually against, and how to promote and demote autonomy safely, per decision class rather than per agent. You will leave able to place your own on-call agents against a working reference and know where to focus next.

  • Wed Oct 28 | Demo stage
  • Steven Karis, Chief Architect