Stably builds AI-native developer tools ranging from an AI-powered end-to-end testing platform, where users create UI tests using agents instead of writing tests manually, to Orca, an open-source agent control center for running coding agents. They're trusted by teams at OpenArt and various enterprises across healthcare, mobility and AI infrastructure.
In Stably, every test is an agent navigating a UI in parallel, so browser infrastructure is foundational to the product. The platform runs two distinct test workloads: batch testing, where large suites execute in the background (think CI pipelines), and interactive test authoring, where a user and an AI agent collaborate inside a live browser session to create and fix tests together.
Problem: When slow browsers make tests flaky
Before Kernel, Stably self-hosted headless Chromium browsers on AWS. For batch testing in CI, this worked well enough. Nobody was watching, and boot-up time didn't matter much. But the interactive test authoring mode had a completely different set of product requirements.
In that mode, the AI agent controls the browser while the user watches and actively directs it. The human and agent are co-piloting. That's why browsers must be headful. The user needs a live GUI of the browser to see what the agent is doing in real time. But early attempts at self-hosting headful browsers on AWS were taking over 10, sometimes almost 20, seconds to start. For a user sitting in a live session waiting to collaborate with an AI agent, that delay wasn’t a great experience. Hosting browsers on AWS was affordable, but latency was the dealbreaker.
The deeper problem was runtime performance. Many of Stably's customers run compute-intensive web applications. On resource-constrained browser instances, pages that should render in seconds took a minute or longer. Most test frameworks wait up to X seconds for an element to appear before declaring it missing. On a slow browser, a perfectly healthy application would trip those timeouts and the test would register as a false negative, eroding developers’ trust in the tests.
“You quickly lose trust if you can’t distinguish between a legitimate test failure and flakiness from the infrastructure.”
Jinjing Liang, Co-founder & CEO, Stably
With many users running parallel test suites requiring a burst of 200 browsers or more, all these problems compounded. Stably needed cloud browsers that were headful, with low startup latency, and low runtime latency.
Solution: Choosing Kernel for headful, low-latency browser infrastructure
They needed three things: headful, fast cold starts, and consistently low runtime latency. They started the search by evaluating other managed browser infrastructure providers, including Kernel, alongside their existing self-hosted AWS setup.
Many options either could not meet Stably’s latency requirements for their customers’ compute‑intensive applications or did not offer headful browsers needed for the collaborative AI test authoring experience. By contrast, Kernel stood out for its headful browsers, Unikernel architecture, which offers 30 ms cold starts, and a low-latency live view for real-time interaction. With all three requirements met, Stably began migrating to Kernel.
“One agent on a single VM might need 200 browsers. The scaling requirements for each are fundamentally different, and Kernel lets us scale browsers independently without worrying about performance.”
Jinjing Liang, Co-founder & CEO, Stably
Results: 333x faster cold starts, 0 infra-caused test flakiness
Today, Stably runs their whole AI testing platform on top of Kernel. Browser cold starts that took 10-20 seconds on AWS are now 30 ms. Users can easily spin up hundreds of parallel headful browser sessions for test suites without worrying about performance. The sessions are fully ephemeral: spin up, run the test, tear down.
Moreover, all the test flakiness problems are gone. If a test fails, it's a real application bug, not an infrastructure issue. And with Kernel handling all browser infrastructure, Stably's engineers focus entirely on their core AI testing platform instead of managing browsers.
“It's been great not having to be an expert on managing headful Chromium instances. We love Kernel.”
Jinjing Liang, Co-founder & CEO, Stably
If you're building a testing platform and unreliable browser infrastructure is getting in the way, let us fix that.
