Started with an AI agent, now doing a thing that won't scale

Vivek Agrawal shares his startup journey, which began with the idea of creating an AI agent to automate manual QA testing for web applications. Initially inspired by AutoGPT, he recognized the potential of AI in simplifying QA processes. Despite lacking knowledge in AI and QA testing, he immersed himself in learning about end-to-end (E2E) testing and the challenges it presents, such as flakiness and maintenance difficulties. Agrawal developed a proof-of-concept using GPT-3.5, but faced limitations with its performance. After exploring various AI models and tools, he pivoted to using GPT-4 Vision to create a minimum viable product (MVP) that could identify interactive elements on web pages. He then engaged with a YC-funded company, GimBooks, to gather feedback and understand their QA needs. Realizing the unreliability of his AI agent, he opted to manually write E2E tests using Playwright, embracing the concept of doing things that won’t scale to provide immediate value to his users. This experience reinforced the importance of early user engagement and understanding their challenges to refine his product.

- Vivek Agrawal's startup focuses on automating QA testing using AI.

- He initially struggled with AI and QA concepts but learned through research and collaboration.

- The MVP was developed using GPT-4 Vision to identify web elements for testing.

- Engaging with users early helped refine the product based on real needs.

- Agrawal emphasized the value of doing non-scalable tasks to provide immediate benefits to users.