In 2023, McKinsey set up a lab and timed more than 40 of its own software developers on routine tasks, with and without generative AI tools. With the tools, they documented code in 50% of the usual time, wrote new code in slightly under 50% and refactored existing code in nearly 2/3 of the time.
In 2023, McKinsey set up a lab and timed more than 40 of its own software developers on routine tasks, with and without generative AI tools. With the tools, they documented code in 50% of the usual time, wrote new code in slightly under 50% and refactored existing code in nearly 2/3 of the time. On high-complexity tasks the saving shrank to less than 10%. For a business paying for a website, that split matters, because some parts of every project resemble the routine tasks and others resemble the complex ones.
Developer Labor Costs
Most of a website budget pays for people’s time. The US Bureau of Labor Statistics put the median annual wage for web developers at $90,930 in May 2024, which works out to about $44 an hour before benefits or an agency’s margin. Every hour removed from a project is worth at least that much, and usually more once overhead is added. The same agency projects 7% employment growth for web developers between 2024 and 2034.
Those hours are of different kinds, and they respond differently to AI. Writing page templates, connecting a contact form, producing documentation and fixing layout bugs are routine. Designing a custom pricing tool or connecting the site to an inventory system involves decisions the developer still has to make.
Coding Assistants on Routine Work
Coding assistants suggest whole functions as a developer types, explain unfamiliar code and convert a plain description into a first draft of a component. The McKinsey results show where that helps most. On a project with 120 developer hours, of which 60 are routine code and documentation, cutting those 60 hours by 50% saves about 30 hours, or roughly $1,300 at the median wage.
Stack Overflow’s 2025 Developer Survey found that 84% of respondents use or plan to use AI tools in their work, and 51% of professional developers use them every day. A business hiring a developer in 2026 is likely paying for someone who already works this way, and it can ask how the quote reflects that.
Matching the Tool to the Project
The largest saving comes from not hiring a developer for work that does not need one. A brochure site with a few pages and a contact form has no custom logic. For that kind of project, an AI website builder produces the layout and first draft of the copy, and the owner pays a subscription with no developer invoice at all.
The calculation changes once the site needs features a template cannot provide. A booking calendar tied to staff schedules or a members’ area requires code. In those cases AI lowers the cost of the developer’s routine hours, while the hours spent on design decisions stay roughly the same.
Accuracy Review
Every suggestion from an assistant has to be read before it is kept. In the same Stack Overflow survey, 46% of developers said they did not trust the accuracy of AI output, against 33% who did. The most common frustration, cited by 66%, was AI solutions that are almost right without being fully correct.
A suggestion that is almost right can take longer to debug than code written from scratch, especially when the error is subtle. McKinsey reported a related finding about job tenure. Developers with less than a year on the job sometimes took 7% to 10% longer with the tools than without them, because they could not yet judge which suggestions to keep.
Handover Documentation
Documentation is the task McKinsey found AI cut most cleanly, to 50% of the usual time. When it is skipped, the business is left dependent on the original developer. A site delivered with notes on how each template works and how to update the plugins can be handed to a different developer without a costly rediscovery phase.
For a small business, this saving appears later. When the original contractor is unavailable, the next one starts from written notes, and the first invoice covers work on the site in place of hours spent reading through it.
Testing Before Launch
A developer can ask an assistant to write automated test cases for a checkout form, covering empty fields and invalid card numbers. Writing those tests by hand is slow, and a site without them can break the first time a customer enters something unexpected.
A test that checks the wrong thing passes without catching anything. The developer still has to read each test and confirm that it checks real behavior.
Written Content for Each Page
Writing the words for a site is a separate cost from building it, and AI reduces both. A developer or owner can generate first drafts of service descriptions and the short alternative text that screen readers announce for each image. Alternative text is a common gap on small sites because writing it by hand for a large product catalog is slow, and a generated first pass turns the job into one of checking.
An image description that names the wrong product or color misleads the customers who rely on it most, so every generated description should be read against the photo it describes before the site goes live.
Questions for a Developer’s Quote
When collecting quotes, a business can ask which parts of the build the developer expects to speed up with AI and how that shows in the price. It can also ask how generated code is reviewed and tested, and if the handover will include written documentation.
A developer who uses AI well should answer those questions plainly. Vague answers, or a quote that charges full hourly rates for template work, are a reason to compare a second quote before signing.
Limits on the Savings
The savings have a ceiling set by the share of routine work in the project. A site that is mostly custom logic, connected to payment systems or customer records, gains little, since the McKinsey figure for complex tasks was under 10%. Security review against the OWASP Top 10 list and decisions about how the site should behave remain human work, and cutting those hours to match the routine savings would trade a smaller invoice for a larger risk.
The 10% Figure
Routine coding and documentation can take about 50% of the usual time with AI, so a business that wants a lower development bill should push as much of its website as possible into that category through templates and standard features. Custom development should be kept for the few functions that require it, and budgeted at close to full price, because the measured saving on complex work was under 10%.
Last updated on
