If you have followed the debate about AI and coding, you have seen two headlines that seem impossible to reconcile.
In a 2023 controlled experiment, developers using GitHub Copilot built an HTTP server in JavaScript 55.8% faster than those without it (Peng et al.). Then, in 2025, a randomised trial by METR found that experienced open-source developers working on their own large projects were 19% slower with AI tools, even though they expected to be 24% faster and still believed afterwards that they had been sped up (METR).
Both studies are careful. Both are probably right. The difference between them is the most useful thing a web team can understand about AI.
Why the results differ
The Copilot study measured a self-contained, well-specified task that many developers had not done before. That is the ideal case for AI: lots of boilerplate, a clear goal and little hidden context.
The METR study measured experienced people working in codebases they knew deeply, with high standards for quality. There, AI output often needed checking and correcting, and the developers already knew most of what the AI was suggesting.
For a web team, the lesson is simple. AI helps most with unfamiliar, routine or clearly defined work, and least with nuanced changes to complex systems you already understand well.
Adoption is high; trust is not
Stack Overflow’s 2025 survey found 84% of developers use or plan to use AI tools, up from 76% the previous year. Yet 46% said they distrust the accuracy of AI output, compared with 33% who trust it, and the top frustration was “AI solutions that are almost right, but not quite” (Stack Overflow).
“Almost right” is expensive. A bug that looks correct takes longer to find than an obvious one.
AI amplifies the team you already have
Google’s 2025 DORA report, based on research with software professionals around the world, reported AI adoption at around 90%, and found that AI was linked to higher delivery throughput but also to more instability. Its central conclusion was that AI acts as an amplifier: strong engineering practices turn AI speed into better delivery, while weak ones make existing problems worse (DORA 2025).
DORA highlighted capabilities such as working in small batches, strong version control, a clear AI stance and a user-centric focus. None of these are about the AI tool itself.
Where AI genuinely helps on web projects
In day-to-day website and front-end work, these are the tasks where AI consistently saves time:
- Scaffolding new components, pages and templates from a clear spec.
- Boilerplate: form validation, API calls, schema markup, build configuration.
- Tests: drafting unit and end-to-end tests that a developer then reviews.
- Refactoring suggestions for legacy code, reviewed carefully.
- Documentation and comments, especially for handover.
- Translating between platforms, for example turning a static component into a HubSpot module or a WordPress block.
- Explaining unfamiliar code during a migration or audit.
Where experienced developers must stay in charge
- Architecture and platform choices.
- Security: authentication, data handling, third-party scripts.
- Performance budgets and Core Web Vitals.
- Accessibility, which generated code frequently gets wrong.
- Final review of every change before it merges.
A practical team policy
- Approve specific tools and define what code and data can be shared with them.
- Require human review of all AI-assisted changes, with the same standards as any other code.
- Keep changes small and well tested, so problems are easy to spot and roll back.
- Measure outcomes, such as lead time, defect rate and performance, not lines of code.
The takeaway
AI coding assistants are neither magic nor hype. They are a fast, occasionally wrong junior collaborator. In the hands of a senior team with good habits, they remove a lot of routine work. Without those habits, they produce more code, faster, and more to fix later.
That is how we use AI in our website development work. For the testing side of the same idea, read AI-assisted website QA.
For the bigger picture on vibe coding, and whether AI will replace web developers, read why developers still matter in the age of AI.
