Maven Clinic Says Real Users Reshaped Its AI Assistant After a 10x Usage Surge
hugobowne · x · 2026-07-25
In a Maven Clinic podcast conversation, the team explains how real users changed the roadmap for its production AI assistant.
Key points
- The assistant launched with only 20% of users able to access it at first.
- Maven originally used it to help people find providers and book appointments, but 50%–60% of early conversations turned out to be basic health questions.
- Four months later, the assistant is available to everyone, benefits answering has shipped, and weekly conversations have grown about 10x.
- Production usage exposed where the team needed better guardrails, better source documents, and better evals.
Why it mattered
- Real user behavior forced the team to rethink what the assistant should do.
- Failures in production became new tests.
- Better models exposed broken eval assumptions rather than simply making the product look better.
The post says the podcast includes a discussion with William about what changed and what he would build differently now.
More from Companies & People
- OpenAI, Anthropic and Microsoft appear on a broad AI industry signatories list — koltregaskes · 2026-07-25
- Jason says frontier model vendors will keep competing with their biggest customers — markjeffrey · 2026-07-25
- New blog post explains how to train a frontier-level world model — dejavucoder · 2026-07-25
- Gabriel says he defaults to open and free models, but is not firmly pro-open yet — gabriel1 · 2026-07-25
- Microagi says it uses Alchemy to manage Cloudflare, AWS, GCP, and K8s as IaC — samgoodwin89 · 2026-07-25
- A new take says OpenAI’s PM-heavy culture makes it feel behind despite frontier progress — ctjlewis · 2026-07-25