Cedar Innovation & Technology Culture

What is the unique story that you feel your company has with AI? If you were writing about it, what would the title of your blog be?
Cedar’s AI story is one that’s familiar to all of us — the patients who face the nearly universal experience of receiving healthcare and managing what comes next: the bill. To deeply understand these needs, my team built a large language model-powered pipeline that transcribes and parses thousands of patient call recordings to gather the insights and data that fuel our AI voice agent, Kora. Kora provides patients with comprehensive and compassionate medical billing support, delivering personalized guidance on everything from balance inquiries to payment options — even when patients can’t fully articulate their problem.

Our blog title would be: “The AI That Listens When You Don’t Know What to Say.” Healthcare billing is an area where users are often confused, anxious and unsure of the exact problem they are facing. It’s why building AI for healthcare is different from typical consumer applications. Our patients aren’t looking for a specific product code or a simple return process — they need guidance, diagnosis and a path to resolution.

What was a monumental moment for your team when it comes to your work with AI?
One monumental moment occurred last year, when we released our AI chat assistant and were overwhelmed by positive feedback from the users of the product — call center agents who provided billing support to patients. We were excited to have the chance to speak with the call center agents during our over-the-shoulder sessions, where we observed how these individuals were using our new AI tools to supercharge their workflows.

We knew that we had something special when an agent exclaimed, “That’s exactly what we need. I couldn’t have said it better myself!” upon seeing an AI-generated suggestion to a patient inquiry. The qualitative feedback from call center agents and patients, along with the numbers that illustrated the efficiency gains brought by our solution, gave us the conviction we needed to double down on Cedar’s AI investment and make AI central to how we solve patient problems.

AI is a constantly evolving field. Very few people coming into these roles have years of experience to pull from. Explain what continuous learning looks like on your team. How do you learn from one another and collaborate?

Experience is helpful, but we also value persistence, creativity and problem-solving. Our strong team of applied AI scientists not only have deep expertise in AI, but many of them also have exposure to different aspects of healthcare billing, such as insurance claims data. That knowledge, paired with a drive to solve complex problems for patients and clients, is invaluable. We also welcomed a PhD fellow earlier this year, recognizing the importance of fresh perspectives. We’re also now hiring for a staff applied AI scientist.

What sets our team apart is our commitment to learning and collaboration. In our recent prompt tuning retro, we reflected on our experimentation process, surfaced insights and pain points, and outlined action items around tooling, workflows and architecture exploration.

Our weekly brainstorms are another cornerstone — providing space to share ideas and feedback at every stage, from experiment design to results review. This spirit of collaboration is what drives us to solve healthcare’s toughest AI challenges.

Describe a project you’re especially eager to tackle in the new year.

In the new year, I’m excited to keep advancing Kora, our AI agent for patient billing support. Many people still prefer the phone when they have questions about their medical bills, and those calls often happen at stressful moments. We’ve already seen how Kora can make these conversations clearer and faster by authenticating callers, gathering the right context and resolving questions that map to cases AI can safely handle. The team and I are going to build on that foundation and continue improving how Kora prepares information for human agents so they can focus on the more complex cases. For me, this project is about giving patients quick, accurate answers, expanding Kora’s capability to respond to increasingly difficult and varied questions, and helping agents get the clarity they need to support someone well.

What technologies and/or practices is your team leveraging to tackle this project?

We’re building on the combination of high-quality AI models and Cedar’s deep infrastructure of integrations, billing logic and patient-specific data. We can’t script every answer because people ask follow-up questions that take calls in many directions, so we rely on AI systems that can stay within the right data and guidelines while responding naturally. To ensure safety, we use simulation, where Kora practices with an AI acting as the patient; perform a human review of selected calls; follow deterministic metrics that show how people move through the flow of a call; and employ AI-driven issue detection that flags outliers. Recently, we shifted toward giving the AI more freedom in phrasing, as long as it stays accurate, and we’ve already seen more patients completing flows and saying, “That’s all I needed.”

How does this project tie into larger company goals?

This work directly supports Cedar’s mission to empower us all to easily and affordably pursue the care we need. By letting Kora resolve appropriate questions and prepare the essential context before handing off to an agent, we’ve reduced handle times up to 25 percent, even with the complexity of healthcare calls. As more providers come online and call volumes grow, continuing to mature Kora helps Cedar deliver clear, dependable, patient-centered support at scale with the level of accuracy people deserve when dealing with their healthcare bills.

What tools support your day-to-day work?

Three or four years ago, most of our tooling was about helping developers write code faster, manually. That problem has shifted completely. Now the question is: How do you best direct coding agents to do work well?

At Cedar, we've moved from GitHub Copilot to Cursor to Claude Code as our primary agentic development platform and what I appreciate is that Cedar hasn't chased tools for the sake of it. There's a term called "token maxing" — basically pushing developers to use as much AI as possible regardless of whether it's actually productive. Cedar has never done that. But they've also never been slow to move when something genuinely better comes along.

Beyond the coding tools themselves, context has become just as important as the code. We use a tool called Glean to search across our codebase, Slack messages and Google Workspace. Before I spin up an agent on a new feature, I'll use Glean to surface product requirements, old design docs, Confluence threads — anything that explains the thinking behind an existing product. I can come up to speed on something as though I've worked on it for years.

How does your team experiment?

We experiment with new technology in a few structured ways and one of the most valuable is Maker Innovation Day — a monthly dedicated day where anyone in product, design, or engineering can go explore a new idea or investigate a new tool. The only rule is that you share what you learn.

I've taken almost every one of my Maker Innovation days and hyper-focused on agentic development. In that space I've built internal MCP servers that connect our development workflows to our Snowflake databases and I've built coding harnesses that let Claude Code investigate the database, investigate multiple repos, generate full project plans and then actually implement them. That kind of infrastructure doesn't get built during a sprint. It needs a dedicated day and permission to take a real risk.

The AI Guild came out of that same energy. We formed it about a year ago and from it spawned a monthly AI Coding Workshop — no slides, show working code. Walk people through using Cursor, then Claude Code, then MCP servers. Real steps, live in front of everyone. That spawned Slack channels, follow-up conversations and a whole internal community. It has meaningfully expanded who can ship at Cedar.

How does your company adapt to change?

Cedar's approach to change is simple: they listen to the people actually doing the work. Kora, our conversational AI voice agent for medical billing, didn't come from a top-down mandate. They got a team of engineers and product people together, let them canvas the market and trusted them to land on a modern approach. While most companies were still running touch-tone phone trees, Cedar shipped an AI-first voice agent. That happened because they gave the team the room to find the best solution rather than dictating one.

It’s been the same pattern to evolve our development tooling. Cedar runs multi-month pilot programs, includes the developers who do the work, throws out what isn't landing and adopts what is. That flywheel is how we have adopted different tools without losing momentum.

What I find most interesting is how this shapes team dynamics. The Venn diagram of designers, product managers and engineers at Cedar has been collapsing. Designers are shipping code updates through pull request review. Product managers are closer to technical work than ever. We're all in each other's work in a way that lets us move much faster because Cedar was willing to invest in the tools and the culture that made it possible.