Commits on GitHub have doubled in four months, but verification capacity has not doubled alongside them. That gap sits beneath the bright story of AI-assisted software development, where more code can move from an idea to a working feature in a matter of seconds.
The velocity chart looks great, the demos land, and the tools are working. Engineering teams adopted AI coding tools, watched output climb and celebrated. Yet the same rise in output creates a harder question: can testing keep pace with the code being produced?
The speed of AI coding changes the balance
AI coding demos make the new workflow easy to picture. “Someone types a sentence, and a working feature shows up 90 seconds later,” said Mudit Singh, Co-Founder and Head of Growth at TestMu AI. The promise is simple: a person describes what they want, and an AI system turns that sentence into a working feature.
That speed helps explain why CTOs and VPs of engineering describe achieving fast, confident adoption and a strong first quarter with AI coding tools. When a team can produce more features in less time, the increase in commits offers a clear sign that the tools are changing how engineers work.
But commits measure activity, not correctness. The numbers show that code is arriving faster, while the ability to verify it has not doubled in the same period. That leaves teams with a growing amount of output to check, even when the demos look successful.
“The bugs started climbing right after the AI rollout, and nobody wants to say it out loud,” said Anna Meadows, CTO of CodeROI. Her warning points to the problem hidden behind a strong velocity chart: working demonstrations can coexist with bugs that climb after AI tools enter the development process.
When the same AI writes code and tests
AI can write a feature, generate the tests for it and explain why those tests passed. That creates a smooth development loop, but it also raises a concern about where the tests get their assumptions.
“If the same AI system creates both the code and the tests, it may repeat the same assumptions in both,” said Margarita Simonova, founder of ILoveMyQA.com. A test can pass because it checks the logic the AI already used, not because it proves that the feature works outside those assumptions.
That distinction matters. Tests may pass because they validate the logic the AI already used, not because they independently prove correctness. If the code and the tests share the same mistaken idea, a passing result can hide the mistake instead of exposing it.
The issue is not that the tools fail to produce useful work. The velocity chart looks great, the demos land, and working features can appear 90 seconds after a sentence is typed. The issue is that verification has not kept up with the number of commits produced during the same four months.
This creates a tension for teams adopting AI coding tools. They can celebrate more output and a strong first quarter while still facing a larger testing burden. Fast, confident adoption does not remove the need to check whether tests challenge the code or simply repeat its logic.
The central warning is easy to miss when every demo succeeds: more code requires enough verification to establish that the code is correct. With commits on GitHub doubling in four months and verification capacity holding below that pace, the gap is now part of the AI coding story.
AI tools can write features, create tests and explain passing results. Teams still need to recognize the difference between a test that passes and a test that proves something. That difference becomes harder to ignore when bugs start climbing after the AI rollout.
Based on
- Commits on GitHub have doubled in four months. Verification capacity has not. — thenewstack.io
- How Your AI Writes Code Faster Than Your Team Can Trust It — forbes.com
- SDD Fixed How AI Writes Code—Nobody’s Fixed How Humans Verify It — forbes.com
- When AI Writes Both The Code And The Tests, Who Checks The Assumptions? — forbes.com
