AICon Beijing 2026 opens call for speaker proposals
AICon, a Chinese AI development and application conference, has opened its call for speakers for the Beijing edition on December 18–19, 2026 at the Crowne Plaza Beijing Sun Palace, with 12 tracks spanning inference infrastructure, agent engineering, agent evaluation and production reliability, agent security, AgentOps, self-evolving agents, AI product design and overseas product expansion. Zhang Ming, Jiang Peng and Zhou Jun are co-chairs, with Yang Jie, Cheng Yan, Chen Shu, Yu Yao, Liu Zhongbing and Zheng Yangfei as track producers. Submissions should be 40-minute talks grounded in real production systems, including metrics such as task success rate, cost and latency; product-pitch talks are explicitly excluded.
It is one of China's larger practitioner-focused AI engineering events, and its track list offers a signal of the field's shift from raw model capability toward production-grade agent engineering.
Agent 会不会写代码，已经不是 2026 年开发者最关心的问题。
更难的问题是：它能不能在真实系统中持续工作？出了错能不能定位？权限能不能收住？成本能不能算清？生成的代码能不能维护？一个能在演示中完成任务的 Agent，距离真正进入生产环境，还有多少工程问题要解决？
过去一年，大模型的能力继续向推理、工具调用、多模态和长任务延伸，开发重心也随之发生变化。团队开始从单次调用模型，转向设计 Agent 的上下文、工具、执行环境和反馈回路；从关注一次回答是否正确，转向评估完整任务轨迹、异常恢复和长期稳定性；从追求更强的单一模型，转向根据质量、延迟和成本进行多模型调度。