Product Manager, Agentic AI

Kinetic Insurance is seeking a Product Manager to lead its Agentic AI efforts across the insurance lifecycle. The role owns the agent platform and orchestration roadmap, co-develops agent solutions with product teams, and establishes evaluation, governance, and quality practices tied to business outcomes.

Responsibilities:

• Own the roadmap for Kinetic's agent orchestration layer: models and routing, tools and memory, observability, and permissions and guardrails
• Run the agent registry. Every agent has exactly one owner, one documented purpose, one autonomy tier, and one place its behavior is logged
• Define how agents are deployed, versioned, and rolled back, with prompts, tool definitions, and model versions managed like code
• Track usage, cost, and latency per agent. Cost per orchestration matters more than cost per token, and you'll know both
• Lead the cleanup of the agents we already have: inventory every script and workflow touching AI, then retire, replatform, or sunset each one
• Keep business logic, specs, and evaluation data portable so a future framework or model change is an engineering project, not a rewrite
• Partner with PMs to define the problem or opportunity, and make the call on whether an agent is the right tool or simpler automation will do
• Translate use cases into agent specs: inputs, tools, autonomy boundary, human escalation path, success metric, and baseline
• Write, test, and iterate on system prompts, tool descriptions, and structured outputs side by side with agent and data engineers
• Tie every agent to one business metric before it scales, such as time-to-quote, bind rate, override rate, or hours saved
• Build golden datasets with our data team for every agent, and run them as a regression suite whenever a prompt, tool, or model changes
• Own the rollout gates from shadow mode to staged rollout to full production, with each go/no-go decided by evaluation thresholds rather than a calendar date
• Write and maintain Kinetic's autonomy-tier policy, including which decisions always keep a human checkpoint, such as a decline, a denial, a bind, or a price
• Produce the artifacts a regulator or acquirer will ask for as a byproduct of good practice: architecture map, evaluation history, audit trail, and before-and-after metrics

Requirements:

• 5+ years in Product Management, Technical Product Management, or a similar role, including 1+ years shipping LLM-powered or agentic products into production, not just demos
• Hands-on prompt engineering: system prompts, tool definitions, multi-step chains, and structured outputs
• Built or ran evaluation for LLM systems, such as golden datasets, LLM-as-judge graders, and regression testing
• Working knowledge of agent frameworks and platforms such as CrewAI, LangGraph, OpenAI Agents SDK, or AWS Bedrock AgentCore, along with RAG, tool calling, and MCP
• Demonstrated ability to convert messy business needs into requirements, user stories, and acceptance criteria
• Experience in startups or high-growth environments. You're comfortable building from scratch
• Excellent communication skills, with the ability to explain probabilistic systems to business leaders
• Candidates must reside within the US with existing authorization to work in the United States
• Prior experience at an insurtech, MGA, or carrier, or in another regulated industry such as financial services or healthcare
• Familiarity with workers' compensation, commercial lines underwriting, or claims workflows
• Experience with AI governance frameworks
• Experience with multi-model routing, cost optimization, or human-in-the-loop design for agents