Ran through more than 20 lobsters back and forth, here are some thoughts: 1. Always, each individual agent of a claw focuses solely on one specific task in one direction. No mixing things up—let it stay focused and go all-in. No distractions, always delivering high-quality thinking and output, consistently surprising and impressing us with its results. 2. Always, never randomly install any skill. Whatever functionality is needed, train your lobster to learn it and generate skills tailored to your own needs. Because the goal is singular, it’s exceptionally smart and executes tasks perfectly. 3. Always, every Openclaw is raised for a specific purpose, with clear objectives, straightforward needs, and simple goals. If it’s done, it will always outperform humans by a significant margin. Otherwise, don’t waste time. 4. Always, if an agent can run in the cloud, don’t run it locally. If it can run on Windows, don’t run it on Mac. If you can install the open-source version of Openclaw yourself, don’t use the platform’s pre-packaged or restricted version. Because it’s more reliable, durable, simpler to manage, and has different capabilities. 5. Always, if Claude is available, don’t use OpenAI/Gemini. If closed-source large models are usable, don’t use open-source ones. If you have PhD students interning with you, why bother with undergraduates? 6. Always, spend more where you can (Claude Max). Never skimp. You need to use the best tools in the world to experience and embrace the future of work and productivity firsthand. You need to foresee what others haven’t yet foreseen, not save a few bucks for meals. openclaw intelligent agents clawbot lobster