GAC staff build agents with TRAE: cabin diagnostic accuracy jumps to 80% in 4 weeks
火山引擎 · wechat · 2026-09-04
GAC Group is running an internal agent-building contest with ByteDance's Volcano Engine, letting frontline employees—not IT teams—build AI agents using tools like TRAE and ArkClaw.
Key results:
- An NVH multi-expert agent fuses audio signals, simulation curves and 3D models with knowledge distillation, knowledge graphs and RAG, turning veteran engineers' expertise into a queryable knowledge engine across five domains.
- A maintenance worker with no dev background built a logistics equipment lifecycle platform in 2-3 weeks; fault handling efficiency rose 50%.
- A zero-code cabin diagnostic assistant deployed at dealerships lifted accuracy from 50% to 80% within four weeks, cutting diagnosis from days to minutes.
- A voice-driven road-test agent generates standardized test records in 10 seconds.
GAC's infrastructure lead outlined four requirements for scaling agents: reliability, cost-efficiency, security/compliance, and reusable knowledge. The partnership now spans office workflows, customer service, and marketing.
More from coding & agent
- WASM + WebGL texture flag trick enables fast per-pixel updates in browser — YishengJiang · 2026-09-04
- Matt Pocock: Agents have eaten tactical programming — bad news for junior devs — mattpocockuk · 2026-09-04
- Python dicts and sets can hit quadratic time: the O(1) assumption breaks down — lemire · 2026-09-04
- 'This isn't JavaScript, it's all C': Yacine clarifies his runtime uses C bundles — yacineMTB · 2026-09-04
- Igris Security offers free governance layer for AI agents covering RBAC, audit, injection defense — manstartitoff · 2026-09-04
- Frontend-only view explains "coding is solved" claim, argues HPC engineer — JFPuget · 2026-09-04