At 3 AM, I found myself still writing a "resume" for my AI. During that time, I felt a bit ridiculous. I kept copying and pasting my WeChat official account positioning, writing style, and product structure into different AI chat boxes. Each time, I had to start with, "Hello, I'm Xiao G, and my style is..." Then I'd close the chat and start over.
A month later, it hit me:** I wasn't using AI; I was just being its assistant.**
So, I spent some time testing mainstream AI knowledge base tools on the market—like Feishu, Notion, IMA, and NotebookLM. The conclusion? Each has its pros and cons, but most suffer from two major issues: either the AI can't access all of my notes, or it can't directly manipulate my local files.
In the end, I opted for the Obsidian + Claude Code combination. The reason was simple: it allows AI to truly integrate into my workflow. I gathered all my materials—hundreds of drafts, recordings, and product documents—into one place. The AI can see everything and can directly modify the content.
It's worth noting that the AI models powering this system aren't fixed. I use Claude for writing, switch to Midjourney for images, and might employ Kimi for processing long audio. If I had to apply for separate APIs for each model and handle different calling formats, efficiency would plummet.
That's where aggregation platforms like GPTProto come in. To put it simply, they consolidate multiple large model APIs into a single entry point. The workflow I built in Claude Code can flexibly call the most suitable model behind the scenes without major overhauls. This keeps costs manageable and stability assured.
The results were immediate.
The most noticeable change was with my WeChat account. Previously, writing a few thousand words would take an entire day. Now, from topic selection to final draft, it takes about an hour. The quality hasn't diminished; in fact, the metrics have improved. The biggest shift from monthly to daily updates isn’t about working harder; it’s that the tools have made daily updates feasible.
I've also streamlined my Xiaohongshu posts, consistently producing around 20 notes a day. Some may not get high views, but they still convert. The AI has learned my tone and style, creating content that resonates with readers.
What truly made me feel like I had "arrived" were two key achievements.
**First, I created two virtual products for sale—one priced at 39 and the other at 599. **From concept to copy and detail pages, all of it was completed through the collaboration of AI modules in my knowledge base. It acts not just as a writer, but also takes on some of the responsibilities of a product manager.
Second, I attended an in-person class and fed several hours of recordings into the system. By the end of the afternoon, it had helped me organize over a dozen core methodology documents. When I write industry analyses, I can directly cite specific insights from the instructors. Knowledge transformed from a static archive into a resource readily available for use.
That’s the key. Many people use AI as a search engine, asking one question at a time. But my approach turns AI into a "digital partner." I predefine collaboration principles, personal guidelines, and target directions in the configuration file. In the workflow folder, I outline the execution steps for various tasks, like the complete process for "writing a Xiaohongshu note."
It doesn’t need to ask, "Who am I?" every time. It knows.
I also implemented an "iteration mechanism": have the AI generate two versions of the copy—one for me to tweak manually, and the other to remain unchanged. Then, I provide feedback by comparing the modified version with the original. After a few rounds, the content it generates gets closer to my desired style. This isn’t just about fine-tuning a tool; it’s about training an assistant that understands me better.
Of course, this method has its challenges. It requires a certain level of file management skills and some time to set up the initial framework. The early stages are an investment, not a quick-fix solution.
But once it’s up and running, the results are sustainable.
Now, I spend less than a third of the time I used to on content creation. I use the saved time for thinking, collaborating, and developing new products. A friend’s ten-person team generates less monthly revenue than I do alone. It’s not because I’m exceptionally talented; it’s because I found the right leverage.
The purpose of tools isn’t to make us busier; it’s to give us more freedom.
That person who was up at 3 AM writing a "resume" for AI can finally shut down the computer and get some sleep.
All-In-One AI APIs and AI Models for Developers | GPT Proto
Access powerful AI APIs and models with GPT Proto, including OpenAI, Claude, Gemini, and more. A fast, scalable, and cost-effective platform for developers.
gptproto.com
Top comments (0)