AI code share shows how much of the committed code was generated or substantially assisted by AI. Code turnover shows how much merged code was reverted, deleted, or substantially rewritten within a defined period. When AI code share increases alongside turnover, engineering leaders have a pattern worth investigating. It isn’t proof that AI caused the rework..
Key Takeaways
- In one Larridin customer environment, AI code share reached 56% while 30-day code turnover rose 64.7% and PR cycle time rose 138.7% during the same period. The figures don’t establish causation, but the combination warrants investigation.
- In a separate customer environment, human-only code had a 15.16% 30-day revert rate, compared with 0.2% for AI-assisted code. Human-only code reverted about 76x as often, showing why organizations need to measure their own results rather than assume AI code is inherently more or less durable.
- AI code share is a composition metric, while code turnover is a quality signal. Tracking them together helps leaders determine whether a growing AI contribution is associated with durable work, additional rework, or a broader change affecting the team.
What AI Code Share and Code Turnover Measure
AI code share measures the percentage of committed code that was generated or substantially assisted by AI tools. It answers one specific question: How much of the code entering the codebase came from AI?
It doesn’t show whether that code was useful, durable, secure, or cost-effective.
Code turnover measures the percentage of merged code that is reverted, deleted, or substantially rewritten within a defined period, typically 30 or 90 days after merge.
A 30-day window captures changes made relatively soon after merge. A 90-day window captures later revisions that the shorter period may miss. Neither metric explains why the code changed.
Together, the measures provide more context than either one alone. Code share shows the composition of the work. Turnover shows how much of the merged code changed again.
What the Customer Pattern Tells Us
In one Larridin customer environment, AI code share reached 56% while 30-day code turnover rose 64.7%. PR cycle time increased 138.7% during the same period.
Those figures describe three concurrent changes:
- AI contributed a larger share of committed code.
- More recently merged code was changed again within 30 days.
- Pull requests took longer to move through the delivery process.
The pattern raises useful questions, but it doesn’t answer them. AI-assisted code may have contributed to the increase in turnover, or the changes may reflect a broader shift in project type, team composition, release pressure, requirements, or review practices.
The organization needs to segment the data before deciding what happened.
Revert Rate and Code Turnover Are Different Signals
A revert rate tracks code or changes that were rolled back. Code turnover is broader. It can include reversions, deletions, and substantial rewrites.
That distinction matters when interpreting a separate Larridin customer finding. In that environment, AI-assisted code had a 0.2% 30-day revert rate, compared with 15.16% for human-only code. Human-only code reverted about 76x as often.
The finding doesn’t prove that AI-assisted code was 76x more durable overall because it doesn’t include every type of turnover. It does show that AI-assisted code was reverted much less frequently in that specific environment during the measurement window.
Taken together, the two customer examples make the same management point: aggregate assumptions about AI code quality are unreliable. Results can vary significantly by organization, team, workflow, tool, and type of work.
Why Higher Turnover Requires Context
Rising turnover may signal avoidable rework, instability, or code that failed to meet the original need. It can also reflect changing requirements, planned refactoring, feature evolution, or routine maintenance.
That’s why turnover should be broken down by:
- AI-assisted versus human-written code
- Team, repository, and service
- Feature work, maintenance, bug fixes, and infrastructure changes
- Pull request size and complexity
- Coding tool and workflow
- Reason for the change
- 30-day versus 90-day measurement window
The goal is to identify where turnover is unusually high, whether it is concentrated in a particular type of work, and what is driving it.
How to Interpret Rising AI Code Share
Higher AI code share isn’t inherently good or bad. It means AI contributed more of the committed code.
That contribution becomes meaningful only when leaders pair it with quality, delivery, cost, and outcome measures. A team can have high AI code share and low turnover, suggesting that AI-assisted work is holding up well. Another team can have the same code share and much higher turnover, requiring a closer look at task scope, review, architecture, or workflow design.
Larridin’s measurement framework makes this distinction explicit: AI code share should be paired with code turnover and other quality signals rather than treated as a performance goal by itself.
The question isn’t whether AI generated more code. It's whether the resulting work delivered value and held up long enough to justify the investment.
What Engineering Leaders Should Investigate
When AI code share increases alongside code turnover, leaders should ask:
- Where is the turnover concentrated? Is it limited to one team, repository, service, or type of change?
- Which code is changing? Is AI-assisted code turning over faster than comparable human-written code?
- Why was it changed? Was the original code defective, incomplete, obsolete, or revised because the requirement changed?
- What happened during review? Did reviewers have enough context to verify intent, architecture, tests, and risk?
- How large were the changes? Are oversized or complex AI-assisted pull requests associated with more turnover?
- What did the work cost? Did the durable outcome justify the resources spent on generation, review, testing, and later revision?
The answers determine the response. One team may need clearer task scope. Another may need smaller changes, stronger review context, better testing, or limits on where AI-generated code can be used.
Frequently Asked Questions
What is code turnover?
Code turnover is the percentage of merged code that is reverted, deleted, or substantially rewritten within a defined period, typically 30 or 90 days after merge.
Is code turnover the same as code churn?
No. Code churn broadly describes modification activity. Code turnover focuses on recently merged code that is reverted, removed, or substantially rewritten during a set measurement window.
Does higher turnover mean AI-generated code is creating technical debt?
Not automatically. Higher turnover is a signal that more code is changing again after merge. Leaders need to determine whether the cause is poor quality, changing requirements, planned refactoring, workflow problems, or another factor.
How do we compare AI-assisted and human-written code?
Measure turnover using version-control history, then use available tool telemetry, commit metadata, developer reporting, or other source signals to segment the code. Document work whose origin can’t be identified reliably rather than forcing false precision.
How can teams reduce avoidable code turnover?
Investigate the cause first. Depending on the pattern, teams may test clearer task scope, smaller changes, stronger review context, better validation, or workflow-specific quality controls. Continue measuring turnover to see whether the change improves durability.
See Whether Your AI-Assisted Code Is Holding Up
Larridin’s AI Dev Productivity platform measures AI code share alongside code turnover and other delivery signals, helping engineering leaders see whether greater AI contribution is associated with durable work or additional rework.
Book a discovery call to understand how AI-assisted code is performing across your teams and repositories.