Flint is an autonomous website platform that generates custom, on-brand landing pages. Think of it as a marketing agent working behind the scenes to get your message out to the world. They're trusted by Cognition, Graphite, 11x, Tandem, and countless others to grow their revenue.
Under the hood, Flint's agent relies heavily on browser infrastructure to navigate to a customer's site and clone their brand. It's a web-first product, and so the performance and reliability of its browser layer matters a lot.
Problem: Self-hosting infra is a pain
Before Kernel, Flint was running its own Chromium browsers using Docker containers and classic Playwright. For the early days, this approach was fine. Latency wasn’t too much of a concern, and an over-provisioning helped smoothen things over.
As Flint grew beyond the initial launch, the team started feeling the pain of managing their own fleet. Things were slow and about 5% of the sites they needed to access were being blocked by bot detection, and they knew that if they wanted to keep scaling, it was time to find a better solution.
“We didn't want to be running a fleet of Chromium instances. We didn't want to be rolling our own captcha solvers, residential proxies, or custom patches and everything required to make this reliable.”
Sam Sheridan, Founding Engineer, Flint
Solution: Selecting Kernel over self-hosting
The Flint team’s criteria were straightforward: they wanted something that would drop cleanly into their existing Playwright workflows, start browsers very quickly to unlock synchronous customer experiences, and just work.
Our sub-30ms cold starts and Playwright Execution features were both particular standouts. Not only would the built-in Playwright Execution API fit seamlessly into their workflow, but it would also remove the need to include Playwright as a package dependency in their agent. On top of that, our Stealth Mode would abstract away the complexity of captcha solvers, residential proxies—all in a single line of code. After a bit of testing, it was a no-brainer for Sam and the team to start the migration.
“After switching to Kernel, a whole suite of problems just went away.”
Sam Sheridan, Founding Engineer, Flint
Results: 5x fewer bot detection errors, 167x faster cold starts
Today, Kernel powers two distinct sides of Flint’s platform. On the brand-importing side, the workflow remains the same, but bot detection errors—which previously affected roughly five percent of site visits—have dropped to under one percent since switching to Kernel.
But the real unlock came from the performance difference. After switching to Kernel’s browsers, cold starts dropped from 5 seconds to about 30 ms. This enabled an entirely new use case: Flint customers could now run real-time, open-ended web research—visiting competitor sites, taking screenshots for inspiration, and running QA loops.
“Kernel’s reliability and developer experience made it an easy choice to use across more of our platform.”
Sam Sheridan, Founding Engineer, Flint
If you're building a web-first agent, let us help. Your agent deserves better.
