Many people ask what OpenClaw can be used for. Some people use it to make money, write programs, do automation, and take on projects. My own words are mainly used to improve the workflow of running the company. I dismantled different channels on Discord, inserting different system prompts and skills for each channel, corresponding to different projects. Write articles, write code, and study the market. More complex projects will inject the path of the project folder and dedicated Meory library into the system prompt to reduce token waste caused by tool usage. Then I linked all the company's Slack, Linear, GitHub repo, and Telegram groups for membership groups. AI will regularly scan the chat records of member groups. If someone reports a bug or requests a feature, it will automatically determine the severity and directly issue a ticket on Linear to the corresponding person. Whisper can also be used to convert weekly meeting recordings into verbatim transcripts and throw them in to generate meeting summaries and action items. But to make these run, there is a prerequisite: you must first pour in the basic information. The most interesting thing is that when AI has enough context, it can start doing things that you haven't designed before. It knows what was decided in last week's meeting, which tickets were assigned to whom, and which things have been delayed. So when a task exceeds the expected time, it will proactively suggest that I have the PM follow up on the progress. I didn't set a rule for it to do this. It is its own judgment based on the context to do this thing. This is the power of context. The more complete the information you feed it, the more things it can do for you, and many of them are beyond your imagination. In addition to company management, I recently opened a new channel category that has nothing to do with work, which is parenting. My son has just been born, and I want to seriously study how to take care of children. But parenting information is too mixed, and the quality of content in the Chinese community is uneven, with many copied from each other. So I asked OpenClaw to crawl some high-yield parenting blogs abroad and organize the more systematic content sources. Then use NotebookLM's Skill to throw all of this knowledge into it and ask it to output a structured summary file. I created a Knowledge Base using these summaries. Now if I have any parenting questions, just ask them directly on Discord's channel, and I can ensure that there are as few hallucinations as possible, because they are all based on high-quality sources that I have screened. None of the above uses were planned by me before installation. When I installed OpenClaw, I didn't know I would use it to manage tickets. I don't know if I will use it to track the progress of the PM. I don't know if I would use it to study how to take care of children. Every scene is a pain point encountered after installation, and then thinking 'Hey, it seems like it can be used to solve it'. The moment of 'hey, it seems like you can give it a try' is only possible for those who have tools in their hands. People who are not pretending will not even have this idea. The attitude of most people towards AI tools is like this: first think about what you want to do, and then decide whether or not to install it. But this logic has a fatal problem: if you don't touch it, you don't even know what it can do. If you don't actually play with it, your cognitive understanding of this tool will stay in other people's descriptions, screenshots, and tweets. Those are all second-hand information. The biggest problem with second-hand information is that it can only tell you the parts that others find useful. But what really changes your way of working is often something you accidentally discover while playing around. No one would write those things as teaching because they are too personal and only you will know when you encounter them yourself. When I was researching how to use crayfish in company operations, I found that there was no information at all because everyone was exploring. Lobster was born in November last year, and it really became popular in mid January this year, which means it has only been a little over a month since it entered the public eye. All users are pioneers, and everyone is crossing the river by feeling the stones. In other words, most of the emerging use cases did not expect to do so at first. Currently, 90% of people's views on AI still remain at the level of "GPT 3.5 era" chatbots, but the tech industry has long been turned upside down. Today, everyone is discussing Context Engineering. If you haven't encountered AI agents, you won't understand. Tomorrow everyone is talking about multi-agent workflow, and if you haven't run it, you won't understand. Last year, there was a period of time when everyone was copying various prompt templates. Later, a group of people were studying MCP, and now everyone is discussing Skills. The wind direction changes every few months, and without heavy use, you have no idea what these things are doing, let alone why everyone jumps from one to the next. Every moment you choose to 'wait and see' is widening the gap. And this gap has a terrible characteristic: you can't feel it happening. Because you don't know what you don't know. You think you just haven't installed a tool yet. But in reality, what you missed was a whole layer of cognitive update. Those who have used it have a different way of thinking about problems. When they see a task, the path of 'this can be done by AI' automatically comes to their mind. This is not a gap in knowledge, it is a gap in thinking patterns. Knowledge can be supplemented, but thinking patterns cannot. Thinking patterns can only grow through experience. So why do we have to use it for what? The first generation of people who played the internet didn't do it because they knew it could be used for e-commerce and gaming, they just played it because they thought it was cool. In this era of rapid technological iteration and no standard answer, the five words' act first, then act 'may be the most underestimated code of conduct.