GenAI Code Quality Survey

Codurance is a global software consultancy
that assists businesses improving their technical capability to support growth.

We understand that no two organisations are the same. Our adaptable approach enables us to focus support on where you need it most. We work alongside your team, improving your platforms, processes and capabilities, creating modern solutions that are sustainable, flexible and scalable.

Do this 14-question self-assessment and receive a first report highlighting strengths and weaknesses of your GenAI adoption.

genai_adoption_code_quality_light_1784732841361-removebg-preview

How are you using GenAI?

How are you ensuring code quality?

How does AI influence operating cost / team efficiency?

Contact Details

How are you using GenAI?

AI Code Generation Percentage

What percentage of your team's code is currently generated by AI assistants (e.g. Claude, Copilot, Cursor, ...)?

Less than 25%

25% to 50%

Over 50%

We don't know / not tracking this metric

Standardization of AI Tooling

How standardized and collaborative is your team's GenAI developer toolset and configuration?

Centrally managed, version-controlled custom prompt rules (e.g., .cursorrules) and standard settings, with active knowledge sharing via AI workshops or community

Individual developers configure their own custom templates and prompts, sharing tips informally on Slack or wiki pages

Default out-of-the-box assistant configurations with no shared templates or collaboration guidelines

Developers use whichever AI tools they see fit

AI Usage Guidance and Guardrails

What guidance, training, or guardrails do developers (especially juniors) have when using AI coding assistants?

Mandatory training on prompting safety, license checks, and critical code evaluation, plus mandatory pairing for junior AI PRs

Structured guidelines and pair-programming sessions, senior engineers do extra deep code reviews

General guidance policy with no formal supervision, normal code reviews

Especially junior developers are more productive with AI tools, we trust them to use the tools appropriately

Data Privacy and IP Protection

How does your organization enforce data privacy and prevent intellectual property (IP) leakage when developers use GenAI tools?

Enforced enterprise accounts with zero-data retention policies, and automated scanning for IP leaks

Structured corporate policy limiting LLM prompts to non-sensitive elements, but no automated scans

Basic developer guidelines with no formal monitoring or enterprise-grade privacy contracts

Developers can use AI in any way they see fit, we trust them to use the tools appropriately

How are you ensuring code quality?

Test-Driven Development Practices

Does your team mandate Test-Driven Development (TDD) or write unit tests before prompting the AI?

Yes, TDD is strictly mandated and unit tests are written first

Developers write tests but not in TDD style

Unit tests are generated by AI after the production code is generated

We don't write tests before releasing changes

Security and License Compliance Verification

How do you verify that AI-generated code does not introduce security vulnerabilities or copyleft licenses (like GPL)?

Enforced automated SAST/DAST and license compliance scans in the CI/CD pipeline

Manual checklist verification during peer reviews

Security and/or license scanning performed occasionally

We don't verify / no scanning done

Architectural Boundaries and Duplication Management

How do you enforce architectural boundaries and manage codebase duplication against LLM-generated shortcuts?

Centralized design systems, automated duplication checks, and automated architecture validation in the CI/CD pipeline

Rigorous manual inspection by senior software architects during planned refactoring and code review cycles

Addressed on an ad-hoc, reactive basis when boundary violations or duplicate modules cause issues

As AI makes it easy to refactor the code, we don't worry about architectural boundaries or code duplication

Code Churn Trends

How has your code churn (code rewritten/deleted within 30 days of last commit) changed since adopting AI?

Decreased or stayed stable

Increased slightly

Increased significantly, more code smells and bugs have to be fixed

We do not monitor code churn or refactoring metrics

Codebase Complexity Monitoring

Are you actively monitoring codebase complexity metrics to assess the impact of AI tools?

Tracked automatically in every CI build (e.g. cyclomatic complexity, maintainability index)

Tracked periodically during sprint planning or retrospectives

Analyzed only when technical debt slows down development teams

Code complexity metrics are not tracked at all

How does AI influence operating cost / team efficiency?

Pull Request Cycle Time

How has your average Pull Request (PR) cycle time changed since deploying AI tools?

Decreased, PRs get merged quicker than before

Unchanged

Increased, the review process takes longer than before

We don't know / not tracking this metric

AI-Generated PR Review Process

Have you adapted the code review process for AI-generated Pull Requests?

We enforce strict PR size limits and/or mandate that authors explain the AI's logic or reviewers inquire deeply about the logic.

We use the same standard peer reviews as before.

Reviewers often quickly approve complex, large diffs to avoid blocking the pipeline.

We have relaxed peer review requirements or bypass human reviews for AI-generated code to optimize speed of delivery.

Shared Code Ownership and Maintainability

How does your team ensure shared code ownership and maintainability for systems that are heavily generated by AI?

Mandatory deep code reviews, architecture-sharing walks, and strict self-documenting code standards

We rely on standard peer reviews, but lack formal knowledge sharing for generated components

Code is mainly reviewed by the developer who knows this part of the codebase best

With AI everyone is empowered to change all parts of the codebase and merge the changes themselves

Feature Development vs Debugging Time Ratio

How has the ratio of time spent writing new features versus debugging/troubleshooting code changed since adopting GenAI?

We spend significantly less time troubleshooting because AI-assisted code is backed by comprehensive tests and robust architecture

The ratio is unchanged, we write code faster but spend relatively the same amount of time debugging and testing as before

We write code faster, but spend more time debugging subtle runtime errors, edge cases, regressions, code smells, architecture or quality issues than before.

We did not compare the before and after and can't tell.

AI-Assisted Legacy Code Refactoring

How does your team utilize GenAI tools when refactoring legacy code or modifying existing complex systems?

We require a comprehensive test safety net (e.g., unit & integration tests and/or characterization tests) before prompting, validating changes incrementally

We use AI to refactor under manual guidance and run existing regression tests to verify behavior

Developers use AI to refactor or rewrite legacy files without complete test coverage, relying on manual testing

We avoid refactoring legacy systems using AI, or let AI rewrite legacy components and deploy without validation checks

How are you using GenAI?

AI Code Generation Percentage

Standardization of AI Tooling

AI Usage Guidance and Guardrails

Data Privacy and IP Protection

How are you ensuring code quality?

Test-Driven Development Practices

Security and License Compliance Verification

Architectural Boundaries and Duplication Management

Code Churn Trends

Codebase Complexity Monitoring

How does AI influence operating cost / team efficiency?

Pull Request Cycle Time

AI-Generated PR Review Process

Shared Code Ownership and Maintainability

Feature Development vs Debugging Time Ratio

AI-Assisted Legacy Code Refactoring

Latest Blogs

Contact Us