Automated testing is a necessity of modern development, but unfortunately, a lot of companies still opt to perform end-to-end tests manually. It's not hard to see why: high upfront development costs, flaky tests, and regular rewrites are a hard sell.
Existing tools to automate end-to-end testing come with extremely high costs and the need for dedicated training - considerations that are just not practical for the majority of development teams.
Whether you're a Freelancer, agency, startup, multi-billion-dollar FAANG corporation, it's our mission to bring automated end-to-end testing to everyone.
We work with highly bespoke, highly complex applications that have a lot of moving parts. Existing attempts to automate end-to-end testing required a huge amount of development time and drastically slowed down development velocity when making changes. Leveraging our in-house Machine Learning experience, we set out on a quest to see if we could resolve these issues.
Our earlier attempts revolved around training an AI model to learn how to use applications through trial and error. The idea was that the model would learn the application's user interface, rooting out edge cases and flagging when it could no longer perform previous functionality. Unfortunately, this self-exploration approach couldn't guarantee that we were testing the aspects of the application that really mattered.
Thankfully, with the recent emergence of state-of-the-art language models, we now have access to models that not only understand instructions but also understand the semantic meaning of your HTML. Taking inspiration from userstory based testing tools such as Behat and Cucumber, we adapted our approach to run tests based on simple language instructions and suddenly everything started to fit into place.
Carbonate is just the start; we have plenty of development tools lined up with the same sales pitch: easy, reliable, affordable.
We hope you join us on this journey to make the web a slightly less buggy place.