At TCL Industries’ software engineering centre, 500 CodeBuddy licences are in deep, routine use, covering more than 90% of the core R&D team, with a plan to extend gradually to the full 2,000-person R&D system. Fixing a playback-stutter bug used to take 8 hours; with CodeBuddy it takes 1.5 hours. A full Android-to-iOS feature migration planned for two weeks now takes two days. In the first two months of 2026 the team used up its full-year token budget.
TCL Industries is the consumer business entity formed after TCL’s 2019 asset restructuring, with products covering smart TVs, phones, tablets and commercial devices sold in more than 160 countries. Its software engineering centre, made up of solutions, application development, system technology and quality assurance centres, has nearly 2,000 people. The work is heavy: over a decade of accumulated code, more than a hundred product lines and release requirements that differ by country in regulations, languages, carriers and policies. New hires typically need three months to get hands-on with the system code. A senior engineer’s eight-hour day had only about a quarter of its time creating value, the rest going to reading old code, locating bugs, writing unit tests and switching tools.
Cooperation with Tencent Cloud on CodeBuddy started in March 2025. TCL chose it because it adapts to the existing R&D process, in-house toolchain, coding standards and business systems without forcing a change of the established development model. Usage began with a pilot of a dozen-plus people in code review and then extended to testing, knowledge accumulation and project management, with each step closing its loop and having its return on investment worked out before the next.
TCL moved the process to an AI-native model driven end to end by AI, from requirement understanding through coding and testing to release: AI generates, checks, modifies and verifies code, failed verifications go back to the AI, and humans review at key nodes. In the bug case, engineers give CodeBuddy the logs and source code; the AI marks three suspect points within 5 minutes, the engineer confirms the plan and adjusts it in about 30 minutes, and CodeBuddy compiles, generates unit tests and runs regression automatically. The Android-to-iOS migration relies on CodeBuddy’s cross-file understanding and batch adaptation.
A TV game-engine project requiring 3D modelling, a skill nobody on the team had, would traditionally have needed one senior and three to five mid-level engineers. TCL tried outside cooperation without good results, then delivered it with one mid-level engineer and the original Android engineers using CodeBuddy. Shen Xuesong, application development director, said the team went from not knowing 3D modelling to knowing it very quickly. He expects staff to split into “super individuals” who command AI and “AI trainers” who accumulate business knowledge, and TCL has created a new job family for AI-native talent.