The new Cloud Coach website is officially live. The relaunch reflects where we are today — and where we believe education software is going. Notably, the entire site was built through Vibe Coding: engineers and AI working in deep collaboration, making the path from idea to launch shorter than ever.
Visitors will find a clearer picture of Education Intelligence, our approach to building software that understands education and helps schools understand, decide, act, and learn. The site also opens up our work with schools: real customer stories and case studies that show the results behind the ideas.
What is new
The new site is organized around the questions schools actually ask: what Education Intelligence is, how it works alongside the systems schools already run, and what it looks like in K12 and higher education. From “why generic AI falls short” to an interactive demo that turns a classroom recording into AI understanding and follow-up action, to complete solution pages for K12 and higher education — every section is designed to answer one concrete school question.
The site publishes in Chinese and English in parallel, with page structure, design system, and SEO settings (sitemaps, structured data, page titles and descriptions) all engineered into a repeatable foundation. Our Insights section will keep sharing perspectives on education, AI, and the future of education software — written by the team on the ground.
What is Vibe Coding
Vibe Coding, coined by AI researcher Andrej Karpathy in 2025, describes a way of building software in natural-language collaboration with AI: the developer states intent, constraints, and acceptance criteria clearly, the AI produces the implementation, and the human focuses on judging whether it is right and good — rather than writing every line by hand. What it changes is not a single tool, but the path from idea to implementation to verification itself.
This is remarkably close to how we build education software every day: agree on goals with the school first, let the system implement, and keep humans judging at the critical points. The only difference is that this time, the implementer is AI.
How we practiced it: Claude Code + GLM 5.3
The entire relaunch was built in the Claude Code environment, driven by the GLM 5.3 large language model. Four lessons we took away:
- State intent clearly: the more specific the requirements (page structure, design specs, content standards), the closer the AI’s output lands to expectations — vague instructions never produce precise software.
- Move in small steps and verify continuously: each section is implemented and visually checked on its own, with layout data measured by probes — never accepted on gut feeling.
- Humans own facts and judgment: content standards, customer information, every number and name we publish is confirmed by a person; the AI drafts and executes.
- Turn experience into process: anything repeatable becomes a documented standard (such as our article publishing workflow), so the next run is faster and steadier.
Every bilingual page, the design system, and the publishing workflow behind them were completed this way — including the article you are reading now.
What it means for education software
We don’t just advocate AI — we build software with it. The process gave us first-hand experience of what schools face when introducing AI: how to build trust, how to keep humans in the loop for critical decisions, and how to help AI understand the real context of an industry. That is exactly the problem Education Intelligence is built to solve: making AI understand education, and helping schools understand, decide, act, and learn.
If you are facing an education IT challenge, or want to talk about how AI can land in your school, we would love to hear from you — contact us.
