TAL Education’s Jiuzhang “Lobster” (九章龙虾) Aims to Turn AI From Brain to Hands in Classrooms

Vertical agents meet the classroom

TAL Education (好未来) has rolled out what it calls China’s first education‑vertical intelligent agent, Jiuzhang Longxia (九章龙虾), and with it a student‑facing companion, “Xiaojinglong” (小精龙). The company says the agent moves beyond simple chat: it can plan autonomously, call specialized tools, store long‑term student memory and execute timed tasks. Can an AI really become a 24/7 study partner? TAL’s CTO Tian Mi (田密) argues yes — and it has been reported that the product is already being positioned as a practical assistant for teachers and students alike.

Skills, sandboxing and the tech choices behind the pitch

Jiuzhang Longxia packages subject workflows into “Skill” bundles that handle homework grading, test diagnosis, lesson creation and even dynamic geometry animations to avoid large‑model hallucinations. TAL says it did not adopt the original open‑source OpenClaw framework verbatim, citing concerns about uncontrollable elements; instead it developed an in‑house variant that it claims lowers cost and raises security, including a sandbox mode that restricts risky operations. It has been reported that Xiaojinglong’s standout feature is persistent, multimodal student memory — a mechanism TAL says enables truly personalized exercise plans and targeted remediation.

Moat, market and policy backdrop

TAL frames its advantage as a combination of vertical professional data and long‑standing teacher relationships. The company points to its Jiuzhang AIXue (九章爱学) platform as both a seed user base and a data reservoir for building education‑specific Skills; it has been reported that the platform already hosts scores of education agents. Meanwhile, China’s big tech players — including Baidu (百度), Tencent (腾讯) and ByteDance (字节跳动) — are racing to ship their own “lobster” style agents, raising the question: where will differentiation come from? TAL is betting on domain expertise and proprietary teaching assets rather than purely generalist models.

Geopolitics and the path ahead

This push for domestically tailored intelligent agents unfolds against a broader backdrop: tighter export controls on advanced chips and years of regulatory scrutiny of private tutoring in China have encouraged local firms to build end‑to‑end stacks and safer, industry‑specific deployments. Whether Jiuzhang Longxia can scale classroom impact while navigating policy sensitivities remains to be seen. For now TAL’s pitch is simple and powerful: reduce teachers’ prep time from hours to minutes and give every student a companion that “remembers” them.