Dumping everything on one AI.
I used to do this too. One chat window, asking it everything: write articles, analyze data, check news, schedule tasks. The result? It did everything, but nothing well enough.
Then it hit me: this is just like running a company. You wouldn't have one person handle marketing, finance, tech, and customer service, right? Same logic applies to AI.
Before explaining why I chose cloud APIs, let me clarify what these two mean:
Local Model: Download the entire AI model (program and parameter files) onto your own computer or server, running it on your machine. Common tool: Ollama. Pros: no API fees, high privacy. Cons: requires significant hardware, usually slower.
Cloud API: You don't run the model locally. Instead, you call someone else's service (OpenAI, Anthropic, Google, etc.) through an API key, letting their servers do the computation and send you results. Pros: fast, typically more capable models. Cons: pay-per-use.
I tried. Ran DeepSeek and Qwen through Ollama.
Conclusion: too slow.
Local model inference speed compared to cloud APIs is significantly worse. If your system needs real-time responses (like a PM coordinating multiple members, or a data analyst needing to deliver results before market conditions change), local models simply can't keep up.
But local models aren't completely useless — I kept Qwen for running knowledge base indexing (Embedding). Indexing doesn't need real-time responses; running slower is fine — what matters is it's free. Save the expensive models for tasks requiring real-time reasoning.
More details in Chapters 5 and 7.
My design logic:
Step 1: Start from pain points. What problems do I need to solve? I need crypto intelligence gathering and analysis, plus content production and publishing, plus various development project progress.