Tang Jie: Founder, Not CEO | Generated by AI

Home 2026.07

I should search for current, verifiable information on this rather than rely on memory—and there’s a factual mix-up worth clearing up first.Question: Give an introduction to the life story of Tang Jie, founder of Zhipu AI (Z.ai).

Answer:

First, a clarification on a common mix-up: Tang Jie (唐杰) is the founder and chief scientist of Zhipu, not the CEO. The CEO is Zhang Peng (张鹏), also a Tsinghua alumnus. Zhipu was founded in 2019 as a spin-out from Tsinghua University’s Computer Science Department by Tang Jie and Li Juanzi, both professors at Tsinghua, with Zhang Peng as Chief Executive. Since you asked about the “CEO’s life story” and named Tang Jie, I’ll cover Tang Jie — the technical founder, who is closer to your idol-set (the Karpathy/Wenfeng Liang archetype of the deeply-technical builder).

The shape of his life

Academic root. Tang Jie holds a doctorate in computer science from Tsinghua University, where he is now a professor. He never left academia in the way a typical founder does — Zhipu is a spinout of his lab’s work, not a clean break from it. The company was incubated inside Tsinghua’s Knowledge Engineering Group (KEG), the research lab now directed by his co-founder Li Juanzi.

The AMiner era (pre-LLM). Before large models, Tang built his reputation on knowledge graphs and academic data mining. He invented the academic search engine AMiner — his Weibo bio reportedly lists only two things: Tsinghua professor and founder of AMiner. This is the through-line of his career: knowledge representation and reasoning over structured data, which later fused with pretrained models. Tang’s vision of integrating massive pretrained models with knowledge and reasoning (as seen with the Wudao project) continues to drive Zhipu’s R&D trajectory.

The pivot to large models (2020–2021). In 2019 Zhipu started as a knowledge-graph company that struggled to secure investment as an academic startup without a clear business model; the Zhongguancun science park gave them rent-free office space for three months to get going. Then in 2020 — earlier than most — they began developing the first large-model AI technologies, recognizing their potential, and the GLM architecture work began. In 2022 Zhipu and Tsinghua’s lab jointly debuted GLM-130B, a 130-billion-parameter bilingual model demonstrating China could build GPT-3-class systems — this predates ChatGPT, which is a meaningful detail for understanding his standing.

Mentor to a generation. This is the part most relevant to you. Tang has personally mentored a whole generation of China’s AI leaders, including two of the founders of Moonshot AI, and serves as vice director of the Beijing Academy of Artificial Intelligence (BAAI). His lab is effectively a node that seeded multiple “AI Tiger” companies — DeepLang, Moonshot, and others trace back to KEG.

Character

The most vivid portrait comes from the recent 36Kr profile and Zhang Peng’s own description. Tang takes as a personal rule a line from academician Wang Xuan — that if a scientist appears on TV too often, his scientific career is basically over — so he hardly manages his personal image and stays almost invisible in the team. At Zhipu’s January 8, 2026 Hong Kong IPO, it was Chairman Liu Debing and CEO Zhang Peng who stood beside the gong, while Tang Jie — founder and chief scientist — was hidden in the team, so quiet as to be almost invisible.

Zhang Peng’s read on him: Professor Tang is exceptionally smart but also highly execution-oriented and very passionate — once he’s figured something out, he pushes extremely hard, with great focus.

The recent arc (2024–2026)

The dramatic turn worth knowing: at the end of 2024, when DeepSeek R1 made its splash and the world called it “China’s DeepSeek moment,” Tang — after 13 years of quiet work — judged that the chat paradigm was ending and its marginal returns dropping sharply, and pivoted toward agents and coding. When asked about DeepSeek he reportedly said just three words: “Very shocking” — the disappointment of a decade-long worker overtaken in the narrative by a latecomer.

The bet paid off in coding models. GLM-5.2 put a domestic open-source large model into the global top three for coding for the first time, alongside Claude and OpenAI, displacing Google’s Gemini. When GLM-5 stumbled in February 2026 and HK$70 billion in market value evaporated in a day amid product-mechanism problems, Tang didn’t explain much — he sent a letter, offered a compensation plan, and continued forward. When announcing GLM-5.2, he framed his philosophy as intelligence that should be “open, accessible, and ready to build with” — consistent with Zhipu’s MIT-licensed open-weight strategy since July 2025.

Why he’s relevant to your trajectory

Tang is almost a Chinese counterpart to your idol-set: a researcher who stayed technical, built foundational infrastructure (AMiner → GLM), mentored an ecosystem rather than chasing personal brand, and bet correctly on the agent/coding shift before the market caught up. The “quiet builder who ships and lets results argue” model maps directly onto your own startup-preparation playbook — persuasion through artifacts, not pitches.

References:

Back Donate