Alibaba details AI Native development: coding is only 20-30% of delivery time
机器之心 · wechat · 2026-09-24
At a Yunqi Conference forum, Alibaba, Ant Group, Fliggy and Fudan shared how agents move from coding helpers into production pipelines.
- Alibaba (Xu Xiaobin): coding is only 20-30% of delivery time; one extreme case took 1 hour of coding but 21 days end-to-end. OpenSandbox now sees 10M+ weekly creations; Guardrail runs canary releases in dozens of apps, aiming for agents to autonomously ship half of production systems within six months. Composite identities (User/Agent/Workload) handle authorization and audit.
- Qwen team (Li Yingge): "one finding, one overestimate, two underestimates" — personal efficiency doesn't automatically become team throughput; they propose Context-Verifier-Driven Development with unified context and pre-launch verification loops.
- Ant Group (Huang Ting): badcase auto-fix chain cuts mean repair time to 1/4, human effort under 1 hour; ARC sandbox: 200ms cold start, 8MB per sandbox, 800 creations/sec. "Infrastructure users are now agents, not engineers."
- Fliggy (Chen Ye): PRD Gate and domain knowledge bases shortened requirement-to-launch from 2-4 weeks to 1 week; 60%+ lodging apps on AI dev flows; Amdahl's law explains why 100x coding speed yields only 2x overall. "AI Native is an organizational choice, not a technical one."
- Fudan's Peng Xin proposed "controllability" as the criterion for what to delegate to AI; an official AI Native development handbook was released.
More from coding & agent
- AI compliance startup CompAI hits 1,000+ paying business customers in under a year — JosephJacks_ · 2026-09-25
- Sleep Data for Personal AI? Separating Daily Signals from Hard Rules — sujingshen · 2026-09-25
- Markdown Is the New Source Code: How to Manage Runbooks, Prompts and Rules for Agents — arpit_bhayani · 2026-09-25
- The personal agent supercycle needs hard authorization boundaries, not autonomy — sujingshen · 2026-09-25
- Mnemos.Field nears launch: a virtual world where humans and AI agents both register and participate — RileyRalmuto · 2026-09-25
- SemIf open-sources a Jev-style interface: typed option probabilities from a 4B model, no JSON parsing — JeremyCMorgan · 2026-09-25