Database › ByteDance
ByteDance
No mandate reported. A ByteDance VP says AI's share of the company's code grew more than 6x in a year, but warns against chasing that single metric.[1]
Tools
Metrics tracked
- Over the past year, AI's code contribution rate grew more than 6x, AI-coding token use 5x, and the AI-code merge rate more than 2x.[1]
- TRAE is used by 92% of ByteDance engineers.[2]
Guardrails
- ByteDance found AI-generated output scored only 40-60 out of 100 on usability, reliability and maintainability; adding harness work (context, architecture constraints, team knowledge) raised correctness from about 80% to nearly 90% and deliverability to about 80.[1]
Hiring, reviews & process changes
Nothing verified from a public source yet. Know a source?
Reported results
- The TRAE team had more than 90% of its code written by AI over six months, but per-person throughput rose only 1.6x, which ByteDance cites as evidence that code share is a misleading metric.[1]
Metric tags: % code AI-generated, Adoption rate, Quality / defects, PR throughput
Editor's note: ByteDance sells TRAE, so its internal figures are self-reported by the vendor. The 2026 figures come from a talk by VP of Technology Hong Dingkun.
Sources
- 163.com (IT Times): Hong Dingkun on the challenges behind ByteDance's 6x growth in AI code contribution · 2026-06-24 · press source
- Longbridge: TRAE used by 92% of ByteDance engineers · 2025-12-18 · press source
Last verified 2026-10-25. Industry and size are editorial bands; see methodology. Spot an error? Submit a correction.