Doubao and Qianwen both chose to roll back their AI agent features on July 15, the same day—this is no coincidence, but the result of new regulations forcing the move.

That day marks the official implementation of the "Interim Measures for the Management of Anthropomorphic Interactive AI Services."
The regulatory requirements are clear:
anti-addiction, minor identity verification, and primary responsibility for content review.
On June 26, Shanghai's "Qinglang·AI Application Chaos Crackdown" had already removed 14k non-compliant AI agents, and even MiniMax's "one-click undressing" feature was called out; it was inevitable for platforms to pull back at this point.
But regulation is only the surface; the real driving force is that the business logic doesn't work.
UGC AI agents are a typical high-cost, low-efficiency business:
high frequency, lightweight, low per-unit value.
They consume massive computing power while generating good-looking user activity metrics that can't be monetized.
When the cost and revenue sides are compared, platforms act.
Compare two completely different transformation paths:
- Doubao C→C:
Channeling AI agent traffic to the Catbox App, keeping it on the C-side for handover—essentially migrating the same user base.
- Qianwen C→B:
On June 3, opened Agent and Skill access; Luckin, KFC, Mixue, and China Eastern Airlines have already gone live; cutting C-side UGC is making way for the B-side.
Both paths indicate one thing:
The AI application market has moved from land-grabbing to value validation.
The next competitive dimension will no longer be user scale or feature count, but compliance capability and sustainable revenue models.
If you still want to spend time building your own AI agent, you have only two options:
either go B-side to find real business scenarios, or don't waste your effort on it. 🤠