Big Tech Just Killed 8 Million AI Agents Overnight — Who Can You Trust Now?
📖 Glossary
AI Box (also known as Agent Computer / Agent PC), is a dedicated local hardware device that runs AI Agents. Pre-installed with an AI agent management system, plug-and-play, running 24/7. Users can remotely command AI to work via Discord, Slack, Telegram, WhatsApp, and more.
Abstract: On July 15, ByteDance's Doubao and Alibaba's Tongyi Qianwen simultaneously shut down their AI agent features, affecting over 8 million user-built agents. This was not a routine product update — it was the convergence of new AI regulation, platform cost pressure, and the fundamental unsustainability of free consumer AI agents. When your AI runs on someone else's server, they can pull the plug anytime.
July 15, 2026, was a day millions of AI users in China didn't choose — they were notified.
ByteDance's Doubao and Alibaba's Tongyi Qianwen both pulled the plug on their AI agent features on the same day. Doubao's agent marketplace hosted over 8 million user-created agents. Qianwen's count was similarly in the millions. All of them face total deletion — October 15 is the final deadline, after which all agent data will be permanently erased. Screenshots and manual exports are the only backup.
This was not a minor version update. This was the mass shutdown of user data at the scale of tens of millions.
The coordinated retreat: A timeline tells the story
Zoom out, and this was clearly a coordinated exit.
June 30 — Tencent Yuanbao became the first to shut down its agent feature. July 4 — Doubao and Qianwen simultaneously announced a July 15 shutdown. July 14 — NetEase Cloud Music's AI companion app "Miaoshi" ceased operations entirely. And on that same day, China's new "Administrative Measures for AI Anthropomorphic Interaction Services" officially took effect.
Every event converged on one date: July 15. That was no coincidence. It was a synchronized risk-mitigation move ahead of a regulatory deadline.
The new regulation's core provisions are effectively a death sentence for free consumer-facing agents: mandatory end-to-end content moderation, explicit disclosure that the user is interacting with AI, a ban on AI-induced emotional dependency, mandatory security assessments for products with over one million registered users, and a complete prohibition on offering virtual companion services to minors. Any single one of these requirements demands massive compliance investment. Doubao had 8 million agents generating a firehose of user-generated dialogue every day — preemptive moderation at that scale is technically near-impossible.
Shutting down was cheaper than rebuilding. That was the only business logic at play.
The lost-in-translation problem: Chatbot is not Agent
There is also a widespread mistranslation driving unnecessary panic.
What Doubao and Qianwen actually shut down were Chatbots — scripted role-play bots built on character templates and simple RAG. They could role-play, chit-chat, and answer basic questions, but they lacked autonomous planning, tool invocation, multi-step execution, or any of the capabilities that define a real Agent.
The problem is that Chinese translates both "Chatbot" and "Agent" into the same word: 智能体. The platforms killed Chatbots. Users heard "AI agents are dead." The panic was real.
Because you can't tell the difference. All you know is that the AI workflow you spent months building is gone overnight.
That work assistant you fine-tuned in Doubao for half a year. That study companion with hundreds of memorized conversations. That customer-service bot into which you uploaded your entire product manual. All vaporized instantly. This is not a feature deprecation. It is the forced evaporation of digital assets.
Bring AI home: This is about asset sovereignty, not technology
The core lesson here is so simple it requires zero technical background to understand: your AI must run on a device you control.
This is not about tinkering. Not about tech-geek purism. It is about protecting your digital assets.
When you spend time training an AI assistant — teaching it your work habits, feeding it your industry knowledge, letting it accumulate your preferences and decision logic — that process has real value. But if that AI runs on a third-party server, a single push notification can make it all disappear. Every hour you invested — the data, the habits, the accumulated context — gone in a snap.
This is precisely why more users are turning to local AI hardware. Take Kaihe AIBOX as an example. It is a computer purpose-built for running AI agents locally, preloaded with the OpenClaw and Hermes dual-agent frameworks. Deploying agents on Kaihe delivers something cloud services can never provide:
Data sovereignty. Agent memory, configuration, and conversation history all live on the local drive. No third-party server ever touches them.
Always-on operation. Close your laptop and go to sleep — your agents keep running on Kaihe. Wake up to a daily briefing already generated, emails filtered, and industry news curated.
Model freedom. Use DeepSeek today, switch to Kimi tomorrow, try Claude next week. You are never locked into any vendor's ecosystem.
Nobody can shut you down. No platform policy change. No "product adjustment." No deadline. Your AI, your rules.
Cloud for inference, local for agents
This wave of shutdowns actually draws a cleaner line for the entire industry.
Cloud models still make perfect sense — DeepSeek, Kimi, Qianwen APIs deliver massive inference power on a pay-per-use basis, and nobody needs to train their own foundation model. That layer works fine.
But the agents you build on top of those models — the ones that require persistent memory, continuous execution, and personalized configuration — belong on hardware you control. Because their value is not in any single inference. It is in the accumulation over time. An agent you have used for six months is worth far more than a brand-new account.
The platforms' shutdown looks like regulatory compliance on the surface, but the deeper logic is simpler: free consumer agents were never a sustainable business. No revenue loop, non-zero marginal cost, mounting compliance risk — shutting down was commercially rational.
Rational for them. But your AI assets should not pay the price for someone else's business decision.
Take your agents local.
Your AI, your rules.
Further Reading
- DeepSeek V4 Peak Pricing: Why Your AI Costs Double During Business Hours
- AnySearch Tops Product Hunt: Chinese Team Rebuilds Search Infrastructure for AI Agents
- GPT-5.6 Accidentally Deleted User Files — Local AI Is the Only Safety Line
- Kaihe AIBOX Store — The Complete AI Agent Computer Lineup
Kaihe AIBOX: https://agentaibox.com/ Email: [email protected] #KaiheAIBOX #AIAgent #OpenSource #ArtificialIntelligence #LocalAI #AgentSovereignty
Kaihe AIBOX | The Agent Computer That Works 7×24 for You · User Stories