Kael Zhang
TencentAI ProgrammingCodeBuddyAI Code GenerationSoftware DevelopmentAI Agent

Tencent Executive: Majority of This Year's Code Generated by AI, AI Programming Enters Deep Water

Kael Zhang

June 5, Tencent Cloud AI Industry Application Conference.

In a conversation with Chief AI Scientist Shunyu Yao, Tencent Senior Executive Vice President Dowson Tong dropped a number that silenced the room:

“This year, the majority of Tencent’s code is generated by AI.”

This was not PR speak. Tong elaborated: engineers now delegate coding to AI, spending more time on architecture design, requirements decomposition, and periodically guiding and correcting AI output.

Just eight months earlier, Tencent’s 2025 R&D Big Data Report put this figure at 50%.

From 50% to “majority,” the speed of Tencent’s AI programming evolution is redefining the underlying logic of software development.

From 50% to “Majority”: CodeBuddy’s Penetration Curve

In October 2025, Tencent first systematically disclosed the penetration progress of its AI programming tools:

The phrasing at the time was “for every two lines of code written, one is completed with AI’s help.” Now, AI is no longer an “assistant” but the primary source of code output.

CodeBuddy is built on Tencent’s Hunyuan large model. In 2025, Tencent added an average of 325 million lines of code per month, fulfilling 370,000 business requirements, with R&D personnel accounting for 76% of total staff. With AI support, R&D automation levels increased by 67%, saving 5.3 million manual operations monthly.

Specific business data is even more compelling:

BusinessResults
WeChat BackendCompilation time reduced by 50%
WeChat PayRequirement delivery cycle shortened by 31%, release quality improved by 14%
Mobile QQiOS compilation build time reduced by 40%
Tencent GamesArt production automation rate reached 95%
Tencent Cloud65% of new code from AI, per capita bugs per thousand lines of code reduced by 31.5%

Two Executives’ “Contradictory” Statements

Almost simultaneously, two iconic figures in the tech industry made seemingly contradictory statements.

On June 1, at COMPUTEX Taipei, NVIDIA CEO Jensen Huang, wearing his signature leather jacket, said:

“Saying that AI has reduced jobs is complete nonsense.”

His argument: GitHub code submissions surged from 300 million in 2023 to 1.4 billion in the first few months of 2026, with software engineer numbers increasing rather than decreasing.

Four days later, Tong said: the code is written by AI; humans do architecture design.

Putting the two statements together creates cognitive dissonance for programmers.

But upon closer analysis, they are not contradictory. Huang was talking about “job numbers,” while Tong was talking about “job content.” Engineers have not disappeared, but their work content is being rewritten—from “writing code” to “writing specs + reviewing code.”

Tencent’s internal data confirms this shift: in job descriptions for new hires, “proficiency with AI programming tools” has gone from a bonus to a requirement.

Four Stages of AI Programming

Tencent’s evolution is not an isolated case. Across the industry, AI programming is following a clear penetration curve:

Stage 1: Code Completion (2023-2024)

GitHub Copilot-style single-line/block-level completion. AI is a “faster input method.”

Stage 2: Function Generation (2024-2025)

From comments to function bodies, AI can understand requirements and generate complete logic. Tencent’s CodeBuddy reached 50% penetration at this stage.

Stage 3: Module-Level Development (2025-2026)

AI begins to undertake complete business module development, with humans responsible for interface design and acceptance. This is where Tencent is now.

Stage 4: Full-Stack Autonomy (2026+)

AI covers the entire process from requirements documentation to deployment and operations. The human role further contracts to “product manager + architect.”

Tencent’s “majority of code generated by AI” means it has crossed into Stage 3.

Tencent’s AI Investment: Aggressive and Unconcerned with Short-Term Costs

The rapid improvement in AI programming penetration is backed by massive capital investment.

In 2025, Tencent’s capital expenditure was 79.2 billion yuan, with a large portion going to AI infrastructure. In Q1 2026, capital expenditure soared to over 31.9 billion yuan, a year-on-year increase of 16%, with R&D investment of 22.54 billion yuan, up 19% year-on-year.

UBS predicts Tencent’s 2026 capital expenditure will reach 170 billion yuan. The report notes that the company is adopting an aggressive strategy of “sacrificing short-term profits to ensure comprehensive upgrades of large models and cloud business.”

Tencent President Martin Lau stated at the March earnings meeting that 2025 capital expenditure fell short of expectations due to GPU supply constraints, and if conditions allow this year, AI and model investment will at least double.

This scale of investment and the efficiency of AI programming output form a positive feedback loop: more AI infrastructure investment → stronger model capabilities → higher code generation quality → faster R&D iteration → more business requirements.

Signal Significance for the Industry

The importance of Tencent’s disclosure lies not in its use of AI to write code—numerous companies domestically and internationally are making similar attempts—but in its “scale” and “frankness.”

With tens of thousands of engineers, Tencent is one of the world’s largest R&D teams. When a company of this magnitude announces that “the majority of code is generated by AI,” it means:

  1. AI programming is no longer an experimental project, but a productivity tool validated at massive scale
  2. Code quality is controllable, otherwise hundreds of millions of lines of AI-generated code would bring catastrophic technical debt
  3. The engineer transition path is clear, with the shift from writing code to architecture design already implemented at the organizational level

This has profound implications for the labor structure of the entire software industry. It is not “AI replacing programmers,” but “programmers who can use AI replacing programmers who cannot”—a prediction that began circulating in 2024 and became reality in 2026.

Core Judgments

Tencent’s case provides several verifiable observations:

  1. The penetration ceiling of AI programming is higher than expected: 50% is not the endpoint, and “majority” is not either
  2. Engineer role transformation is an organizational problem, not a technical one. Tencent’s 76% R&D personnel ratio provides room for this transition
  3. Capital investment is a necessary condition, but model capability is the key variable determining AI code quality
  4. Code review is the core quality control link, with AI participating in 94% of reviews meaning human engineers’ supervisory role remains critical

Tong’s statement is essentially a signal: one of the world’s largest R&D teams has transformed AI programming from an “efficiency tool” into a “production mainstay.”

The speed of this transition is much faster than most people predicted.

Sources