Senior Software Engineer, AI

About Tono

Specialty care is hard to access and even harder to run. Patients often wait months, and clinics rely on fragmented tools and manual workflows.

Tono builds software and operations together to make specialty care faster to access and easier to deliver, starting with dermatology.

We’re building a fully autonomous specialty medicine clinic: continuous, real-time care for patients anywhere.

What makes us different is how we build. We run two clinics of our own, a national telehealth practice and an in-person clinic in DUMBO, Brooklyn. Our engineers work side by side with our doctors and care coordinators to build (mostly) every piece of the software from scratch. The long-term goal is for specialty clinics across the country to run on our autonomous EHR, freeing them from the administrative burden that slows care down.

Today we see patients in 33 states, covering more than 80% of the US population.

You will work closely with the co-founders and a team of engineers and physicians (backgrounds include Amazon, Apple, and leading academic medical centers). We are backed by Village Global (backed by Mark Zuckerberg, Jeff Bezos, Michael Dell, chaired by Reid Hoffman), HOF Capital, and Oncology Ventures.

What You'll Do

Own the agents that run Tono's clinic end to end. You'll design, build and ship AI agents that handle real clinical and operational work, such as drafting clinical notes, triaging intake, coordinating care, and automating the admin work that slows doctors down. You'll work in our TypeScript monorepo (Cloudflare DO, tRPC, Prisma, Postgres, Inngest) alongside the engineers, physicians and care coordinators who use what you build.

You'll also build the foundation other engineers build on: shared agent infrastructure, tool and memory abstractions, durable workflows, tracing, and the evaluation harnesses that tell us whether an agent is actually getting better. On any given week you might be:

• Building an agent that turns a patient visit into a structured clinical note a doctor trusts

• Designing evals and labeled datasets from real clinician feedback, then using them to catch regressions before they ship

• Building self-improving loops where physician corrections flow back into prompts, retrieval and fine-tuning

• Adding guardrails, human-in-the-loop review and safe fallbacks so an agent fails gracefully

• Debugging why an agent did something unexpected, using traces, logs and replay

We build AI that has to be reliable in production. That means structured outputs, evaluation loops, monitoring and clear escalation to a human when it matters.

How We Work