ByteDance plans to launch an AI agent comparable to Meta's Muse within weeks, according to Bloomberg, betting that a consumer-facing assistant can extend its lead in AI video while it works to close the gap with American frontier labs.

Co-founder Zhang Yiming laid out the plan to the company's AI team at an all-hands meeting, Bloomberg reported, citing unnamed people familiar with the matter. He told staff ByteDance can challenge Anthropic and OpenAI at the frontier — but reaching that level would take until 2029 or 2030 at the earliest.

For practitioners, the announcement signals another deep-pocketed entrant in the agent race, this one with a distribution machine most rivals cannot match. ByteDance has not said whether it will pair the agent with a physical device.

What does the agent compete with?

Meta introduced Muse in September. The product talks with users by voice, appears on screen as an animated character, and helps them complete a range of chores. Meta is rolling Muse out on its smart glasses and on a forthcoming Muse Charm device, Bloomberg reported.

ByteDance's version would compete directly for the same assistant slot — voice, character, task execution — though Bloomberg said it remains unknown whether the Chinese company intends hardware of its own.

Where does ByteDance actually stand on models?

Zhang, 43, named data as his top concern. He said the training datasets behind ByteDance's Seed large language models sit orders of magnitude below what frontier rivals use, according to Bloomberg.

The gap shows in the record. Seed stands alone among China's top AI players in having neither produced a language model that ranks among the world's best on benchmarks nor released its weights to outside developers. The Seed 2.1 update in June was ByteDance's last major language model release.

Zhang also argued, again, that ByteDance should avoid distillation — training a compact model on the outputs of a larger one — and stay focused on long-range goals. The position carries political weight: White House officials and Anthropic CEO Dario Amodei have criticized Chinese rivals including Alibaba, Moonshot AI and Z.ai for building on top of American models through distillation.

How does video fit the strategy?

Video is where ByteDance holds a genuine lead, Bloomberg reported. Its Seedance models stepped in after OpenAI discontinued Sora, and the company intends to commit up to $70 billion to make Seedance the leading video AI platform in China.

That figure frames the agent launch as a beachhead rather than a standalone bet. The company is funding the push with unusual capacity: in September it secured a $29.6 billion loan from close to three dozen banks — a three-year facility with no collateral. Proceeds are earmarked for AI initiatives outside China, including data centers under development in Southeast Asia.

Who is building it?

The Seed team employs around 1,000 people across research and engineering, led by Wu Yonghui, whom Zhang recruited away from Google DeepMind in 2025. In October, Seed became the first of China's big AI labs to launch a separate unit dedicated to safety and alignment.

A ByteDance spokesperson did not respond to Bloomberg's request for comment.

The practical takeaway for operators: an agent with TikTok-grade distribution arrives within weeks, backed by a company willing to spend $70 billion on its strongest category while admitting its language models trail the frontier — and telling investors not to expect parity before 2029.