Last night, Uncle experienced another familiar AI cycle.

Seeing a Web3 project funding news, I thought I’d use AI to quickly organize: project background, funding scale, team situation, important progress, activity mechanism, and participation methods.
At first, I was quite optimistic.
GPT is so powerful now, it should be done in a few minutes, right?
But when the first version came out, the structure was there, but the information was very messy.
The second version asked it to supplement source data, and it started to become very cautious.
The third version asked it to break down the activity mechanism, and it turned into a promotional article for the project team.
I had to keep adjusting the prompt, asking for details frantically. Suddenly, I felt very absurd—I wasn’t releasing myself with AI; instead, I was increasing my workload.
This might be the real bottleneck most people face when using AI: it’s not that they don’t know what they want, but that they don’t know how to translate their needs into prompts that AI can reliably execute.
👉 So Uncle made a comparison using the same requirement:
“Research Web3 projects that completed funding and disclosed within the last month, and organize basic info, funding scale, team situation, important progress, activity mechanism, and participation methods for me.”
On the left is GPT, on the right is xBubble.
GPT’s response is very familiar: it starts answering, listing projects, and writing information.
It’s usable, but I still have to keep asking: where do the sources come from? Is the funding date accurate? Did the project issue tokens? Are there any omissions in the activity mechanism? Which parts are key, and which are just filler?
This is the most exhausting part of using ordinary AI—many key details require me to verify again.
xBubble responds differently, giving a path selection first based on the requirement:
1️⃣ Crypto Research SOP (75% match, estimated 3-4 minutes)
2️⃣ Search & Answer (18% match, estimated 20 seconds)
3️⃣ Compound Skill-Bubble Computer (7% match, estimated 2-10 minutes)
This point is quite crucial: SOP isn’t just a simple “prompt template,” it’s a pre-run task route.
For such tasks, it first plans how to search, what sources to use, how to cross-verify, and how to structure the output. The system arranges all this for you.
For example, in Web3 project research, ordinary large language models tend to answer first and then respond, with quality depending on the model’s on-the-spot performance.
xBubble first judges the task type, then matches the most suitable path.
So the difference is quite clear:
👉 Doesn’t write prompts + ordinary AI: the model answers first, user refines later.
👉 Writes prompts + ordinary AI: user breaks down needs into a whole workflow.
👉 Doesn’t write prompts + xBubble: the system first helps you decide which route to take.
This is also the idea behind xBubble’s Low-prompt AI Agent.
Users speak naturally, and the system chooses the path.
AI should learn AI.
AI should also use AI.
@dappOS_com