Liang Rubo said on August 6 the company will stick with in-house AI research, even as Doubao, Feishu and Volcano Engine deepen integration and over 90% of new Feishu clients buy AI products.
ByteDance CEO Liang Rubo told employees at the company's 2026 mid-year all-hands meeting on August 6 that its large language models (LLMs, AI systems trained on vast text data) are falling further behind leading overseas rivals, even as the company remains competitive in consumer-facing AI applications and video generation. Liang said Doubao stayed strong in consumer applications in the first half of 2026 and Seedance still holds a State Of The Art (SOTA, top benchmark performance) position in video generation, but described the core LLM race as a more fundamental contest tied to the future of Artificial General Intelligence (AGI, human-level general AI). He said ByteDance will not rely on external models to dress up its products for short-term rankings or market attention, and will instead accept near-term lag while continuing long-term in-house R&D. At the same time, ByteDance is tightening coordination across Doubao, Feishu and Volcano Engine, with Liang saying the integration is already showing results as over 90% of new Feishu clients also buy AI-related products. He framed 2026 around three principles, "high priority, thick trunk, long-term optimization," while keeping AI, information platforms and transaction services as ByteDance's three core business pillars.