# Understand and close the AI quality gap in your codebase. > Kerno Audit analyzes your codebase and its commits, and shows where quality slipped after your team adopted AI agents, in complexity, rework and test coverage. Close the gap, and your team saves tokens and engineering time on redoing work. https://kerno.io/audit/ Kerno Audit analyzes your codebase and its commits, and shows where quality slipped after your team adopted AI agents, in complexity, rework and test coverage. Close the gap, and your team saves tokens and engineering time on redoing work. ```sh npm install -g @kerno/cli kerno audit ``` Free. No credit card required. ## AI is helping developers ship more code, but it is also accelerating complexity that compounds fast. Left unchecked, that complexity will build up silently, in code that is becoming slower and riskier to change. - Each engineer now owns more code. +71% code shipped per engineer. - More paths to understand and test in every change. +50% complexity in new code. - AI commits are revisited more than other commits. +43% follow-up changes on AI-written commits. - Test changes nearly doubled, and bug fixes held at a quarter of commits. 25% of commits are still bug fixes. Source: [The Impact of AI on Code Quality, 2026](https://kerno.io/the-impact-of-ai-on-code-quality-2026/) ## Close the test coverage gaps in your high-churn code, before they turn into regressions See where the code your team changes most has no tests, and add them there before the next change breaks something. - **Most-changed code with no tests.** Your most edited functions and files that no test touches, ranked by edits. - **API routes with no tests.** Every API route in your code, split into tested and untested, month by month. - **How often changes include tests.** The share of production commits that also change a test, before and after your team adopted AI agents. ## Catch rising complexity early, while it's still cheap to fix Every if, loop and case is another path to understand and test. See how fast your new code is adding them, and address it before they pile up. - **Complexity added by new code.** Cyclomatic complexity added per 100 new production lines, month by month, with your AI adoption month marked. - **Functions that became too complex.** Every function a commit pushed past cyclomatic complexity 20, with the commit that did it. - **Where complexity is heading.** Where complexity per 100 new lines is heading over the next 12 to 24 months on your current trend, with a 95% range. ## Spot the code that keeps getting reworked, fix the cause, and save engineering time and tokens See where your team keeps redoing the same work, and whether it is fixing mistakes or building on earlier work. Fix the cause once, instead of paying for it again in engineering time and tokens. - **The code reworked most.** The functions, endpoints and files your team rewrites most, with how many edits came within 30 days of the previous one and how many came from agents. - **Whether each rewrite was a fix.** An AI model reads each rewrite and labels it a repair, which fixes the earlier change, or an extension, which builds on it. - **Share of work spent on rework.** The share of commits that rewrite code from the last 30 days, and how many days pass before a change is first rewritten. ## Find out how much of your AI output is real progress, and how much is wasted time and money See whether your team's extra AI output turns into progress, or into churn that keeps costing engineering time and tokens. - **New code that stays in the codebase.** The share of new lines nobody has rewritten or deleted 90 days later. - **Pull requests merged per engineer.** Merged pull requests per active author each month, before and after your team adopted AI agents. - **How big pull requests have become.** Lines in each merged pull request. Bigger changes are harder to review and riskier to merge. ## Works with your codebase, your stack and your agents Languages: TypeScript, JavaScript, Python, Go, Java, Kotlin, Ruby, PHP, C#, Rust, Scala, Swift, C++, C, Dart. Coding agents it recognises: Claude Code, Cursor, GitHub Copilot, Codex, Devin, Jules, Aider, Windsurf. ## Work with our AI quality engineering experts to audit your codebase and close the quality gap Kerno Audit shows where quality slipped after your team adopted AI agents. Our engineers work with your team to find out why, and what to fix first, so your team spends its time and tokens on the changes that matter most. ### Why teams bring us in - **Quality is slipping in your codebase.** Your agents ship more code than your team can track, and you need to know where to start. - **Rework is eating your tokens and time.** Your engineers and agents keep redoing the same code, and every round burns tokens and engineering time. - **Quality is inconsistent across your team.** Every engineer works with agents in their own way, so the quality of what they ship varies from person to person. ### How it works - **Deep codebase audit.** We run Kerno Audit on your repositories, then our engineers dig into the code behind every hotspot. - **Root-cause analysis.** We work through the findings with your engineers to find why each problem keeps coming back. - **A prioritized plan.** What to stabilize or restructure, where to add tests first, and how your team should work with agents. - **Leadership walkthrough.** We present the findings and the plan to your engineering leads, and answer every question. ### What you get - A full audit of your repositories, with complexity, rework, test gaps and delivery before and after AI adoption - The root cause behind each problem area - A prioritized plan for your codebase and your agent workflows - A walkthrough session with your engineering leads [Book a call](https://meetings-eu1.hubspot.com/ktraiaia/kerno-sales) ## Frequently asked questions ### What does Kerno Audit measure? Untested high-churn code, complexity, rework and delivery, measured from your codebase and its commits and compared before and after your team adopted AI. ### How does it know which commits came from AI? From agent markers in commit authors, co-author trailers and messages. It recognises Claude, Cursor, Copilot, Codex, Devin, Aider, Windsurf and Jules. Commits without a marker count as unknown, so the AI share is a floor. ### Which languages does it read? TypeScript, JavaScript, Python, Go, Java, Kotlin, Ruby, PHP, C#, Rust, Scala, Swift, C++, C and Dart. ### What leaves my machine? By default, the results, short source extracts and the requests to the AI judge. Run with --no-evidence to remove the extracts, or --local to keep everything on your machine. ### Does it measure bugs? It measures rework, including repairs to recent changes. Defect rates come from your bug tracker. ### How many repositories can I audit? Up to 10 per organization. Contact us for more. ### How does this relate to the 2026 report? It runs the same analysis we ran on the 46 codebases in the report.