Free report·
AI in the Developer Workflow
High adoption, uneven trust, and the tasks current tools haven't solved
Code generation is the most common task developers report using AI for, with almost half (49%) of those using AI assistants doing so for this purpose.
The 31st edition of SlashData's Developer Nation survey examines how developers are using AI across their workflows: which tasks they turn to AI for, how much of each task they are willing to hand over, and where the gap between adoption and reliance reveals the limits of current tooling. Drawing on survey data from developers actively using AI-assisted development tools, the report moves beyond headline adoption figures to examine AI use across more than a dozen distinct development tasks. It establishes where AI is genuinely accelerating developer work, where it has found a willing audience for tasks developers are happy to offload, and where capability gaps are suppressing the trust needed for deeper reliance. For organisations evaluating or expanding AI tool adoption, the data offers a lens for calibrating expectations, informing governance, and identifying where human oversight remains essential. For those building AI developer tools, it maps the tasks with high demand and low trust.
Key findings
- Code generation is the most widely used Al-assisted task, with 49% of developers using AI for it, yet fewer than half of those are letting AI handle the majority of the work.
- Editing and modifying existing code has the lowest Al-majority reliance of any high-adoption task, at 40%, despite being the second most popular use case.
- Documentation, code comments, and test generation show the highest reliance scores. Tasks developers are willing to offload and the consequences of Al errors are less immediately visible.
- Where existing workflows and tooling are deeply embedded, such as in DevOps, Al faces its highest adoption barrier. Being capable is not enough; it has to be demonstrably better than systems developers already trust.
- The tasks where Al falls short on reliance (editing, refactoring, understanding, and integrating existing systems) are not peripheral activities. They are central to professional software development and share a common characteristic: AI must work within constraints that already exist.
- Unpredictability is structurally different from other Al limitations. Unlike context or verboseness, it cannot be resolved by scaling, making human oversight in existing codebases a necessity.
Questions this report answers
Which development task do developers most commonly use AI for?
Code generation is the most common task, with almost half (49%) of developers using AI assistants doing so for this purpose. Editing or modifying code and bug detection and fixing follow, both at 44%.
Are developers handing the majority of their coding work over to AI?
No. Less than half of developers using AI-assisted development tools are using AI for the majority of their work, and even those who are rarely hand over full control. For code generation and bug detection or fixing, 46% report using AI for more than half of the task, but three-quarters of those fall in the 51-75% range.
Which task shows the lowest AI-majority reliance?
Editing and modifying existing code shows the lowest AI-majority reliance, at 40%, despite being among the most popular tasks for AI assistance.
Why do developers trust AI less for editing existing code?
AI-assisted coding tools still struggle with three key issues for this task: limited context, a tendency toward excess, and unpredictability. Unpredictability in a mature, interconnected codebase is a meaningful risk that cannot be eliminated by adding context.
Where does AI earn the highest developer reliance?
Generating code comments and annotations, generating documentation, and generating tests score highest on both adoption and reliance, largely because developers are willing to hand off these necessary but unrewarding parts of the job.
Why is AI adoption low for integrating APIs?
Integrating APIs is a task where AI capability is the binding constraint, requiring work across existing systems and live services. Its reliance score of -10% confirms that developers who use AI for it remain firmly in control.
Who it's for
- Organisations evaluating or expanding AI tool adoption across developer workflows
- Teams building governance frameworks for AI-assisted code review and quality assurance
- Builders of AI developer tools mapping where current tooling falls short