A profound realization I had today is: Don't worry about not building anything "particularly grand or valuable" using agents like OpenClaw or Codex 🧐. Simply incorporating AI into your daily life, even if it's just to improve the experience, is already incredibly worthwhile. I've been working on something: developing my own "AlphaGo" to automate the gameplay of Slay the Spire 2. Later, I realized this path wasn't smooth; it was far from truly stable, generalizable, and reliable full automation. But during this process, I suddenly shifted my focus: Since I can't completely "play for me" right now, why not first create "high-quality assistance for me"? The result was a dramatic improvement in the gaming experience 😎. This improvement isn't just about "completing a few more levels." More importantly, the previously tedious, fragmented, and repetitive tasks were significantly reduced and can now be highly customized. For example, the first change: Previously, when getting reward cards, I often had to repeatedly check websites to see how strong a card was and whether it suited my current deck. Now, I have AI create a mod for me, displaying card rating data from designated websites directly after the card name in the game. In other words, the information is directly embedded into the game itself. Previously, I either had to search the web repeatedly or wait for someone else to write the mod. Now I can directly describe my needs and let the agent create it for me. The second change: Previously, many strategies, experiences, and route decisions relied on memorization or temporarily consulting guides and videos. Now, I have AI combine: online experience, existing guides, and our own practical experience to generate strategy suggestions. And this isn't a static strategy library, but a "dedicated external brain" that iterates and upgrades with actual gameplay. Every time I press a shortcut key, it automatically reads in-game information and then provides detailed, customizable suggestions. This is on a completely different level of experience than "searching for a general strategy guide." The third change: Human memory is finite and wears down. Previously, if you didn't play a game for a long time, coming back was essentially like relearning it. You used to have to recall, research, and rebuild your feel for it. Now, you can directly transfer these experiences to AI. As long as you keep clear records of documents and processes, the external brain can basically restore the context for you with a single click when you return. If all these things were done by humans, whether it's recording, reading records, or recovering memories, it's all very tiring. But AI is particularly well-suited for this kind of continuous accumulation, on-demand auxiliary work. I'm increasingly convinced that: the most easily underestimated aspect of agents isn't "whether they can directly complete a large project for you," but rather that they can be seamlessly integrated into various specific scenarios, to enhance tasks that inherently require mental effort. Games are just one example. As long as a scenario includes: information retrieval, contextual understanding, experience integration, real-time judgment, and long-term memory, AI has the potential to bring significant improvements. The key is that the cost of building this system is now incredibly low. If you feel that AI isn't smart enough, the problem often isn't with the model itself, but rather with two things: the requirements aren't clearly described, and the lack of sufficient effective context. AI doesn't read minds; it's more like a highly capable system that needs precise guidance. The clearer your goals and the more solid the context you provide, the higher the quality of its output will generally be. Initially, I also thought AI was a bit confused with specific games like Slay the Spire. Later, I directly fed it several Slay the Spire-related open-source projects, mod websites, and strategy guides, letting it extract, understand, and utilize them. As a result, it quickly grasped the key context and began consistently delivering far more than I expected. Therefore, my current view is: These agents may not yet allow someone with little software engineering experience to easily build large and complex systems. But these technologies are already more than enough to help ordinary people build a layer of truly personalized software around their existing large-scale products, games, and workflows. The barrier to entry is much lower than many people imagine. If you force yourself to "use AI to build a large project" right away, the process is often painful and prone to frustration. It's better to first integrate it into your daily life, and let yourself truly experience a qualitative change in your experience. When you really start to rely on it, you'll be more motivated to continue exploring more complex directions. This may be the most realistic and rewarding path for ordinary people to embrace Agents. Below is a video demonstrating actual usage, built using Codex: