How Felicity builds AI agents to automate healthcare operations with Kernel

Felicity is an AI-native robotic process automation (RPA) platform for healthcare that lets operators quickly build custom workflows across the systems they use every day, without writing code. They are trusted by organizations such as Octave Therapy, Chapter, Thrive Holdings, and Flagler Health, along with many other healthcare providers.

These workflows have to navigate a maze of niche, legacy web portals — from EHRs and payer sites to billing dashboards and scheduling tools. Felicity’s customers automate everything from inbound referral processing and patient chart updates to report generation and provider credentialing. That makes browser infrastructure a foundational part of Felicity’s technology stack.

Problem: You can’t fix what you can’t see

Every workflow Felicity automates touches sensitive systems, so from day one their bar for auditability and reliability was far higher than a typical SaaS product. Their previous browser infrastructure provider tried to meet that bar by storing rrweb events to record and replay the automations, but that approach quickly broke down.

Because rrweb works by recording DOM changes, events, and user inputs and then “replaying” a page by reconstructing the DOM, there is no strong guarantee that a replay will perfectly match what actually happened on screen. This approach cannot reliably capture everything a real browser rendered (for example, content in other tabs, some canvas content, or media assets referenced by a URL that is changed, expired, or otherwise not available), which means it can't provide the reliable audit trail required in the healthcare context.

“We’re dealing with very sensitive systems. You need to see exactly what’s happening. Any time something breaks, you need a high-fidelity replay to debug it.”

Paul Chang, Co-founder & CTO, Felicity

When the replays did work, Paul’s team still ran into reliability issues. Replays would often be cut off at the end, and they failed to load more than half of the time. On top of that, rrweb doesn’t provide first‑class support for workflows that span multiple browser tabs, so any automation that opened new tabs simply fell off the recording. For Felicity, whose automations routinely hop across tabs, this meant they couldn’t see the whole picture when debugging broken customer automations.

They resorted to building their own in-house solution as a workaround. Their solution captured and stitched together screenshots just to reconstruct what an automation actually did. Meanwhile, browser sessions frequently took 5-10 seconds to spin up, making the experience painfully slow for customers. They waited for their provider to fix many of these underlying issues, but the fixes never came, forcing Paul and his team to look for an alternative.

Solution: Selecting Kernel over all others

That experience reshaped what Paul’s team started looking for in an infrastructure provider. Instead of rrweb replays that might or might not load, they needed healthcare‑grade observability and consistently fast performance.

Paul turned to Kernel for our reliable MP4 video replays, first-class multi-tab support, fast performance, and an infrastructure-first team he could trust. Switching providers to test things out was straightforward: they ported existing automations over, validated that our agent identity features worked against their target sites, and discovered their end-to-end runs were now completing 2x faster compared to their previous setup. Plus, they finally had reliable video replays.

With performance doubling and video replays finally working, Paul immediately started migrating their existing automations to Kernel.

“It’s not easy to build serverless browsers that spin up quickly and remain fully observable. Kernel offers exactly what we need.”

Paul Chang, Co-founder & CTO, Felicity

Results: 2x faster automations and healthcare-grade observability

Today, Felicity runs all new browser automations on Kernel. This not only made their customers’ automations 2x faster, but it also eliminated 12 hours/month of incident debugging compared to their previous provider.

“We couldn’t believe how fast the browsers spun up. All of our automations were nearly twice as fast end to end, and the high-fidelity video replays completely changed how we debug broken customer workflows. I cannot overstate the impact Kernel has had.”

Paul Chang, Co-founder & CTO, Felicity

On top of that, a few previously failing automations just worked immediately after the switch.

If you’re currently building process automations and dealing with browser infrastructure, don’t wait — try us out. We promise there is a better way.