Applied AI in practice · 2023
TestForge AI
A Google AI Trailblazers prototype for generating software test scenarios from requirements.
archived
Why it existed
In 2023, Google AI Trailblazers gave teams a short window to test real generative AI use cases. For us, one question was whether LLMs could help generate useful test scenarios from software requirements.
The friction it answered
Writing test scenarios takes a lot of context. A tester has to understand the requirement, the product behaviour, the edge cases, and what has broken before. We wanted to see how much of that first draft could be assisted by an LLM.
What was built
A prototype that used LLMs to turn requirements into draft test scenarios for review. It was built as part of the AI Trailblazers programme, which gave participating teams access to Google Cloud tools and support for building generative AI prototypes.
What it left behind
TestForge AI made the limits of manual context gathering more obvious to me. Test generation only becomes useful when the right surrounding information is pulled in: requirements, recent changes, past failures, related tickets, and the product area being tested. That pointed to a lesson I kept returning to later: AI tools need a way to gather the right context on their own, and the knowledge pipeline is often as important as the AI on top of it.