I would like to share what I learned and felt while creating three AI products, Vrew, Rutton, and LilysAI :) Amazing advances in artificial intelligence are opening up opportunities to ‘automate’ so many human intellectual tasks. But paradoxically, when our team discusses product UX, we often talk about not falling into the ‘trap of automation.’ When you ask AI to do something, it rarely gets 100 points. Most of them end up only scoring around 70 points. Wouldn't it be meaningful because it would save a lot of time just by creating a draft with 70 points? You might think: However, as you follow the user's workflow, you often end up having to destroy the 70-point result created by AI and create 100 from scratch in order to achieve the desired goal. In the case of these types of products, users initially respond by saying, ‘Wow, this is amazing!’, but in reality, they don’t reach a usable level, so retention seems to be low. We call this phenomenon the ‘automatic trap.’ So how should I fill out the remaining 30 points? We were able to solve many issues with the engineering and UX layers. Let’s use our product as an example. LilysAI is a video summary service. The summaries were sometimes inaccurate or overly condensed, raising questions about their reliability. When using existing video summary services, I felt the inconvenience of having to watch the video again because there was no action the user could take when the reliability was in doubt.