Included Health Launches Dot AI Guide With Human Escalation

Included Health launched Dot, an AI-powered healthcare guide, to clients in August. The guide can direct members through care, coverage and administrative questions, but it is designed to pause for a human member care advocate when an agent is uncertain.

A router for healthcare’s branching questions

Dot is part of Included Health’s healthcare-navigation service for employers and health plans. Its main LangGraph workflow, called the Dot supergraph, routes conversations into sub-workflows for urgent-care intake, appointment scheduling, specialist search, behavioral health and other member journeys.

Included Health built the guide as a federated multi-agent system using LangGraph and Deep Agents, with different product teams owning portions of the graph. Rather than duplicate a coverage-answering function inside each journey, the company made it a shared platform sub-agent that multiple workflows can inherit.

Skills supply the service-specific detail

The system represents clinical capabilities and services as skills stored in a virtual filesystem. An agent starts with short descriptions, then progressively loads fuller instructions for the skills it decides fit a member’s needs. Included Health is also working toward encoding third-party employer benefits as skills; it says a plan can contain about 20 to 30 benefits.

The review loop

• Every conversation enters a LangSmith annotation queue for clinical review of care recommendations and emergency-guardrail behavior.
• Those clinical labels are exported for Included Health’s operational metrics dashboards.
• The company used multi-turn simulated-user evaluations while moving four product teams from standard agents to Deep Agents in under two weeks, without significant regressions.

Early performance is based on the company’s own reviews

Included Health reports a 75% lift in chat engagement after launch. It also says clinicians agreed with Dot’s care recommendations above its 95% target in graded conversations, while regular clinical audits found the guide identified more than 99% of high-risk situations.

Those figures come from Included Health’s own review process and are reported in a LangChain case study. The notable design choice is less the number of agents than the attempt to preserve one conversation across automated routing and human intervention—while keeping clinical oversight in the loop.