Lobster that has been tossed around for a while My thoughts: >If you think the lobster is too stupid, remember to turn on the/hint configuration to enable inference, which will consume more tokens but significantly improve efficiency (such as complex inference or code generation) >Claude can effectively replace most scenes, just choose one. Claude is sufficient for most daily scenarios, and the differentiated value of Lobster lies in the flexibility of local deployment and custom agents. I prefer to use Claude to focus on work in front of the computer. Apart from trying new things, most of my lobster usage time is still on the mobile end, which is convenient for quickly handling some problems. >Those who have money and leisure, understand technology, and have special application scenarios are suitable for themselves. Ordinary people don't really need to tinker with it themselves. Gradually, various manufacturers will make optimized versions, which will definitely be easier to use than those that ordinary people configure themselves. I have recently migrated to Clawdi on @ PhalaNetwork, and the debugged version is user-friendly and easy to use. Whether it's prompt tuning, toolchain integration, or operational costs, they will all be better and easier. >You need to think about how to integrate lobsters into your workflow instead of using them to search for scenes. Lobsters are just productivity tools, and heavy tools can actually drag you down. First, sort out the high-frequency pain points in your workflow, and then see what Lobster can solve, rather than the other way around. >From the big picture perspective, the lobster product itself is not the most important. What really matters is that it pushes the concept of personal agent from the developer community to a broader understanding. The future AI interaction is not about "opening web conversations", but rather an agent that resides in the background, has memory, and can operate across applications. The heat will definitely pass, and we need to constantly reflect and iterate our workflow, rather than simply chasing gimmicks >The hype cycle for each new tool is similar: anxiety driven short-term traffic vs. real workflow iteration. The former is a self media business, while the latter generates compound interest. Thinking clearly about what you want to do is more important than being carried away by anxiety. >Mac Mini local big model+lobster, combined with 7 * 24 trading of encrypted assets, can really be a super entity in any corner of the world. Whether it's writing a book, creating a product, or anything you want to do, you can continue to do it. AI has lowered the productivity threshold, while Crypto has lowered the entry threshold for financial infrastructure. The combination of the two has significantly expanded the possibility space for sovereign individuals, and the liberation of sovereign individuals is becoming possible