Case study
Test Flow – Automated AI based E2E Testing Framework
Last updated:
The project at a glance
- The problem
- Explore whether an agent can complete a browser task without a prescribed sequence of clicks.
- My contribution
- Personal project, designed and built solo. Natural-language goals guide browser agents attempting the task.
- Delivery state
- An internal tool I use occasionally alongside other testing methods.
- Result
- Used on Konfig and other projects to surface issues including unclear labels, missing cues and difficult interactions.
First-hand usage observations; not evidence of complete test coverage or measured time savings.
Give an agent a goal, then see whether the interface explains itself.
Conventional browser tests often prescribe a rigid sequence of interactions in code. I wanted to explore a different question: could an agent complete a task when it was told what to achieve, without being told exactly how to use the interface? Test Flow grew from that idea.
Goals and exploration in the browser
I designed and built the tool as a solo project in evenings and weekends. A user connects an AI agent and gives it a job, a sequence of goals or room to explore. The agent attempts the task through the browser experience it encounters. Broader browser support is a direction for the product, while my own browser-based workflow is the current use case.
Why the strongest findings were about usability
The most useful findings were often usability and accessibility problems. When the agent receives no instructions about which button to click, it has to orient itself. Missing cues, unclear labels and difficult interactions become visible in a way that a test with a prewritten click sequence can easily skip over.
I used this approach on Konfig and other projects. In the live configurator, the agent could interact with designs, build its own and exercise editor features. These were behaviours that would have been laborious to cover with conventional point-and-click scripts.
A different lens on quality
Test Flow is an internal tool I use occasionally alongside other testing methods. Its value is the perspective it adds: watching an agent try to understand a real interface can reveal where the product needs to guide its users better. It turns a natural-language goal into a practical exploration of the experience.