NOA: a multi-agent Medicare companion built on Claude

Medicare is notoriously hard to navigate: plans, coverage rules, provider networks, and out-of-pocket costs are scattered across systems that assume a level of fluency most beneficiaries never had a reason to build. For older adults especially, the cost of a wrong turn is real — a missed benefit, an unexpected bill, or care deferred because the path to it was too confusing.

Instead of asking users to become experts, Navigage lets them talk to NOA. The assistant meets people where they are, in plain language, and does the decoding for them — surfacing what a plan covers, who is in-network, and what a given decision will actually cost. Claude Opus 4.8 is the primary model behind that experience.

Under the hood, the architecture is multi-agent. A router agent interprets each user message and dispatches it to specialized sub-agents for costs, coverage, care, reminders, tasks, and coaching. Each sub-agent is equipped with its own tools — connected to plan and provider databases via tool use — so the model can reason over real, current data rather than generic knowledge.

On the text users type, Claude performs several real-time analyses. Intent detection recognizes when someone is asking for a reminder, ending the conversation, or greeting the assistant — replacing brittle keyword heuristics with LLM classifiers that return structured output. The model also extracts clinical and behavioural facts such as date of birth, medications, comorbidities, financial sensitivity, and sentiment; judges confirmations; and estimates out-of-pocket costs for services a plan does not cover.

To keep the experience fast and cost-efficient at scale, Navigage relies on advanced API capabilities — prompt caching, response streaming, structured output, and per-turn effort control. Together they let NOA feel immediate for the user while keeping the economics sound as usage grows.

The result is a real, mature use case of Anthropic's technology as a core product layer. Claude is not a bolt-on FAQ widget here — it is the intelligence that reads intent, holds context, reasons over live data, and turns an intimidating bureaucracy into a conversation an older adult can actually have.