I have been really tortured by AI lately. A few days ago, I wanted to create a workflow that combines automatic data organization and content generation. Nowadays, AI agents, automation, and low code future are being hyped up online every day. At first, I thought this thing should be easy to build and run. After starting to work on it, I directly entered unlimited debugging mode. This prompt is incorrect, please rephrase it. Claude ran out too wide, switch to GPT. The GPT logic is correct, but the format is not working. Then continue changing tools, supplementing rules, and modifying workflows. Sometimes I can't distinguish whether I'm "doing content" or "training AI". And this kind of thing is particularly thought-provoking. The same tool, someone can handle it in just over ten minutes, but sometimes I can callback for two hours based on a result. The high star skills online seem to be getting stronger and stronger one by one, but when put into their own scenarios, many of them simply cannot run smoothly. You copy it, but in the end, you still have to make some changes by yourself. I even had the idea of finding someone specifically to help me with my workflow later on. But upon careful consideration, small teams can actually be quite awkward. Recruiting someone who understands AI is not cheap; Communicating needs, explaining logic, and going back and forth to work can sometimes be even more tiring than tinkering on one's own. Later on, I gradually realized that the biggest problem with AI now is no longer 'not smart enough'. But ordinary people have to learn too much extra in order to make good use of AI. You need to study prompts, model differences, skills, workflows, and which tool is suitable for which task. Real work only takes up a part of the time, and most of the remaining energy is focused on 'making AI work properly'. So when I recently experienced @ dappOS_com's @ xbubble_xyz, I felt that its approach was quite different. Many AI products are now teaching users how to write prompts, build workflows, and adjust agents. But xUbble seems to be doing something else: AI learning AI, AI using AI The biggest feeling I have personally experienced is that there is no need to constantly worry about which model to use for this step. I just need to tell it what I want. Bubble Pilot will recognize the task type on its own and automatically distribute it to the appropriate SOP and execution path. If there is no existing SOP, it will also automatically revert back to a generic agent. The key is that the Bubble Engine in its backend will continue to learn. Which models are suitable for which tasks, which tool combinations are more stable, and which workflows have higher success rates? These previously labor-intensive tasks are now being handled by AI itself. This experience is actually quite enjoyable. Because in the past, it was not that AI couldn't do the work, but that users had already become half programmers themselves in order to make AI do the work. I am particularly impressed by the Bubble Computer mode. Before, when doing a complete task, I had to cut several windows myself: checking materials, organizing, writing content, proofreading, and then outputting. Now it will run the entire link by itself. The local mode, including Bubble Personal, is also quite interesting as it allows users to directly manipulate local files and browsers without having to configure their own environment. I increasingly believe that the truly good AI of the future should not make ordinary people increasingly tired. But rather, AI learns how to use AI on its own. Users only need to tell it the goal, and leave the rest to the system.