Written by: Haotian
As a "shrimp farmer", after enduring weeks of painful torment, I'd like to share a few insights and reflections on shrimp farming, for reference only:
1) Everyone knows shrimp farming is for "efficiency gains", but the truth is, at this stage, most ordinary people's shrimp farming is a "waste of time". There are many unexpected troubles, such as Claude accounts getting banned, API forwarding quotas being blocked, Openclaw suddenly upgrading and "memory" being lost, etc. These consume most of the time, and the meager efficiency gains brought in comparison are far from proportional;
2) Seeing various AI-anxiety-selling Posts on the Twitter timeline, just take a look and move on. Stuff like one command can make Claude control your entire computer, or one Prompt, and the AI will permanently take over your job so you can just lie flat, blah blah blah. The reality is, not only can you not lie flat, you often end up Coding until late at night for a single feature, often fixing one bug only for another to pop up. It's hard to imagine, how those who aren't even willing to spend 5 minutes on actual operation can so righteously shout about AGI arriving and AI overturning everything;
3) Raising lobsters can indeed fulfill many people's OPC (One Person Company) dreams, but the capability ceiling of large models is equal for everyone, yet the cognitive levels of different people in harnessing large models are worlds apart. Don't think everyone can become a Peter Steinberger, Matt Schlicht, or Andrej Karpathy. The so-called shrimp farming ideas, framework design, iteration experience, capability level, and delivery results of a top-notch Dev and an ordinary person can truly be worlds apart;
4) Raising lobsters is essentially building your own dedicated AI OS. Theoretically, the more types of large models connected in parallel, the number of digital employees managed, the hierarchy of active and passive collaborative tasks, the friction level of nesting Skills and combined collaborative task delivery, etc., and the more overlapping demands like Cron, jobs, real-time data Scan, Trading skills, etc., the greater the probability of conflict and the greater the optimization challenge. It's important to know that shrimp farming has never been a "model capability" problem, but an "engineering implementation and optimization" problem.
5) Equipping all digital employees with Opus 4.6 and Gemini 2.5 Flash lite might allow them to do the same things, but the former is like hiring a Wall Street veteran elite, while the latter is more like hiring a black slave from the slums. Maybe they both implement a feature, but the cost consumed and the results delivered are simply incomparable. The truth is, the time you spend struggling to debug, others might easily substitute from the start with "money power". Shrimp farming really burns money. The scary part is, most people understand all this, yet still have to compromise and keep optimizing with inferior models;
6) Managing digital employees is like stacking Legos. The more employees, the more Skills, the more complex the Jobs scenarios, the higher the probability of instantaneous collapse. For certain tasks, sufficient capability is enough; don't be insatiable and impose "demands" beyond your cognitive boundaries. It's recommended to put more effort into memory engineering, Git versioning design, model hallucination removal, etc., otherwise there really will be many moments that push you to the brink of collapse. One second you might be feeling great solving a very Cool requirement, the next second everything collapses and you feel like crying but have no tears. Don't ask me how I know;
That's all.
