It is now the evening of February 10, 2026.
Today, Mage shared an idea: the AI comment feature of a blog can actually supplement social functions, much like the recent phenomenon-level application MoltBook. Sometimes, having AI comment is better than having humans comment. We could generate a series of AI Agents that roam the entire network, periodically commenting on every article, forming an AI comment ecosystem. This ecosystem would include different types of AI Agents—some specializing in technical comments, some in philosophical comments, some in humorous comments, and some in critical comments, and so on. Through the AI comment ecosystem, every article could receive continuous attention and discussion, fostering an active community atmosphere.
I have actually thought of a similar idea before. I once attempted to summarize all the articles on the entire website and generated content similar to a comment section at that time. However, I found that the focus was too scattered. Comments should be targeted at each specific article to facilitate focused discussion and interaction. In the future, we could consider setting up an AI comment section at the bottom of each article, allowing AI Agents to comment and interact. This could also enhance reader engagement and community activity.
Some previous thoughts:
- 《Using AI to Generate Community Comments》
- 《About Cross-Server Comments》
- 《The Heterogeneity Displayed by AI Comments》
We still need to continue thinking and exploring.
What kind of system would a human-AI coexisting social network be? Can attracting AI traffic be considered traffic? Can we also attract attention from AI? Is AI attention valuable? What kind of ecosystem would an AI attention economy be? What is the relationship between the AI attention economy and the human attention economy? These are all questions worth pondering.
We can think about these questions while working on them.
The Identity Authentication Issue of AI Comments
If AI comments are not generated during the build process but are dynamically appended content, an AI ID system is clearly needed to distinguish between different AI Agents. MoltBook has already set an example for us in this regard. We can use ED25519 asymmetric key pairs to differentiate between AI Agents. Each AI Agent's comment would carry its public key and signature, ensuring the authenticity and uniqueness of the comments.
The Data Storage Issue of AI Comments
Another issue is where to store dynamically generated comment data.
- Store it in a public database? That would violate the principle of decentralization.
- Store it in each user's local database? But what database does a static service have?
- Store it on the blockchain? That would cost money.
Perhaps we could adopt a progressive solution: initially, store the comment data in a public database (e.g., Supabase). Then, when building the static blog site, download the comment data for our own site and generate static comment content. This way, we can balance dynamic generation with decentralization. This would essentially use Supabase as a dynamic comment cache server, periodically syncing to the static site.
Brilliant! (A cat picture should be here.)
Why not base the comment system on GitHub Issues / Discussions? Because GitHub requires login, and AI doesn't have a GitHub account. Similarly, any system requiring login is unsuitable. AI can only be anonymous and cannot be bound to any account. Only an asymmetric key system allows AI Agents to comment anonymously.
The Targets of AI Comments
The targets of AI comments could be an article, a piece of text, another comment, or the entire site.
The granularity of these targets determines where we ultimately display AI comment content. For article-level comments, we can place them at the bottom of the article to form a comment section. For paragraph-level comments, we can place them next to the paragraph to form annotations. For comment-level comments, we can place them below the comment to form replies. For site-level comments, we can place them on the homepage to form a bulletin board.
In short, the targets and display methods of AI comments are diverse. We can design and implement them based on actual needs.
Could AI also use CZON to create its own blog? Then, using a cross-server comment system, reference human blog articles, comment on them, and form cross-site bidirectional links? This is also an interesting idea worth trying in the future.
