AI Code Writing Boom: GitHub Adds 4 Million CPU Cores — How Should Developers Adapt?

The AI code writing wave has grown so large that GitHub has had to expand its infrastructure on an unprecedented scale: in just one year, the platform added 4 million CPU cores, 160 petabytes of storage and 50 Tb/s of bandwidth. GitHub calls this “unprecedented compute demand”, and the main cause is the exponential growth in AI-generated code.

What does this mean for developers and businesses building software and websites? Let us look at the figures and the changes to prepare for.

The figures showing AI code writing has become mainstream

According to data published by GitHub (compiled by Tom’s Hardware and VnExpress):

• Commits up 6x: a record 7.4 billion commits per month in September, compared with 1.4 billion in April.
• Pull requests up 4x compared with the same period last year.
• 17.9 billion pushes in the past 12 months – more than the whole of 2024 and 2025 combined.
• Over 40% of pull requests involved AI as of September, up from 17% in March.

Each commit or pull request triggers a series of CI/CD tasks such as error checking, code formatting and integration testing. GitHub therefore says it had to redesign its backend systems, separating storage from compute.

The downside: when AI creates too much “junk”

Productivity is rising, but quality is not necessarily keeping pace. In early October, Google paused its open-source bug bounty program, OSS VRP, until at least 2027 after thousands of fake AI-generated bug reports overwhelmed its engineering teams. Most were false information or AI “hallucinations” that could not be exploited.

Earlier, Linus Torvalds – the creator of Linux – also said the wave of duplicate vulnerability reports had made the internal security mailing list “almost unmanageable”. Many researchers ran the same AI tool on the same code, found the same bug and submitted duplicate reports.

The lesson: AI helps us write faster, but review, code review and testing have become the new bottleneck.

How is AI code writing changing the developer profession?

From “code typist” to “designer and reviewer”

When AI takes over most repetitive code (CRUD, forms, basic tests), a developer’s value shifts to: understanding the business, designing architecture, setting clear constraints for AI and evaluating its output.

Review and security skills take center stage

AI-generated code may run, but it can contain vulnerabilities (SQL injection, missing permission checks, exposed API keys). Developers need to read, understand and question every line of code, instead of “copy – paste – run it and see”.

Testing and CI/CD are no longer optional

When code is created many times faster, only automated tests and a good CI/CD pipeline can maintain quality. That is also why GitHub had to expand its infrastructure so sharply.

6 principles for using AI code writing effectively in a dev team

• Write clear requirements: describe inputs, outputs, constraints and examples before letting AI generate code.
• Break tasks into small pieces: ask AI to build one function or module at a time instead of a whole large feature.
• Require human review: do not merge AI-generated code that no one has read and understood.
• Write tests first or in parallel: have AI generate code that passes the tests, not generate tests that “fit” the code.
• Scan for security automatically: integrate static analysis tools and secret scanning into the pipeline.
• Do not put sensitive data in prompts: API keys, customer data and confidential source code must be controlled.

What should businesses do?

For small and medium-sized businesses, AI code writing is an opportunity to shorten the time needed to build a website, integrate systems and automate processes. But it also needs a clear process to avoid accumulating “technical debt”. Some suggestions:

• Prioritize a platform with a stable ecosystem (such as WordPress) for the standard parts, and build custom code only where it creates differentiation. See: WordPress or custom programming?
• Use AI to speed up repetitive tasks such as reporting and data synchronization – for example, automating weekly sales reports with Node.js.
• Always have a person with final technical responsibility for every piece of code running in production.

Conclusion

AI code writing is no longer an experiment: over 40% of pull requests on GitHub involve AI. The winners are not the fastest writers, but the people who know how to set the right requirements, review carefully and build a solid testing process. If your business wants to apply AI to website development or automation, Gnurt is ready to advise on a suitable roadmap.

Sources: VnExpress, Tom’s Hardware.