AI just got much better at writing code: does that make your app or website cheaper to build?

AI has made the predictable parts of a build faster, but discovery, architecture and judgement still cost what they cost, so whether a cheaper quote reaches you depends on what your project is actually made of.
The short answer
AI has made parts of building software genuinely faster, and in early November 2026 another run of stronger coding models arrived. If you are paying for an app or a website, the honest answer to whether that makes your build cheaper is that it depends on where your project's time actually goes. AI cuts the effort on some parts of a build and barely touches others, so whether your quote falls comes down to how much of your work sits in the parts it helps. This guide splits a build into those two halves, in plain terms, and gives you the one question to ask an agency so any saving reaches you rather than stopping at the keyboard.
What actually happened with AI coding tools this month?
Most of the news this month was about new models and how they score against each other. In early November the companies that make these tools released more capable versions aimed at demanding technical work. Anthropic, for one, put out its most capable coding models to date on 9 November, Claude Fable 5.2 and Claude Mythos 5.2, which it describes as setting a new bar for writing code (Anthropic's announcement). The tech press then measured the new models against one another.
That comparison matters to developers. For a business owner paying for a build, the useful question is a different one. If the tools write code faster, does your quote fall, does the quality rise, and what should you ask for so the saving reaches you? The rest of this guide answers those, without the benchmark noise.
Where does AI genuinely speed up building software?
AI assistants are at their best on the parts of a build that are well understood and repetitive. Boilerplate code, routine tests, data migrations, and first drafts of standard screens are the clearest wins, because the work is predictable and the right answer is largely known in advance. Think of it as a very fast typist who already knows the common patterns by heart.
The measured gains on this kind of work are real. In one controlled study, professional developers finished a self contained task, writing a small web server, 55 percent faster with an AI assistant than without one (GitHub's research). When a job is clear and self contained, the tools shorten it. Much of a straightforward mobile app or software build contains work of exactly this shape, and that is where the time comes out.
Which parts of a build does AI not shorten?
The other half of a build is where the time and the risk really sit, and AI barely touches it. Working out what to build, deciding how the pieces fit together, spotting the edge cases that matter, and checking whether real people can actually use the thing are matters of judgement, not typing.
On large, unfamiliar code the picture gets harder still. A randomised study of experienced developers working in big projects they knew well found they were about 19 percent slower when they used AI tools, not faster (METR's study), because reviewing and correcting the generated code took longer than writing it themselves. The harder part of a building is the plans and the foundations, not laying the bricks a little quicker, and the same is true of software.
Does faster building mean a lower quote for you?
Partly, and in proportion. If a large share of your project is the predictable work above, you may well see the effort, and so the cost, come down. If most of your project is discovery, connecting to other systems, and careful judgement, the saving is small, because those parts were never the slow bit of typing.
It also depends on something you cannot see from outside: how much of the time AI saves goes straight back into reviewing and fixing what it produced. Generated code still has to be read, tested, and made safe, and that work is real. You can get a feel for where your own project's time sits with our project estimator, and our guide to how long real builds take shows which stages tend to dominate a schedule.
How do you ask an agency to pass a saving on?
Here is the one question worth asking, and it does the work of ten. Where in my build does AI save you time, and how does that show up in what I pay? A clear answer points at the predictable parts, boilerplate, standard screens, and routine tests, and shows that saving in the quote or the scope. A vague answer that says the team is simply faster now, with no change to the number or to what you get, tells you the saving is stopping at the keyboard rather than reaching you.
It is also fair to ask where the saved time goes. Put into more testing, better documentation, and a cleaner handover, it buys you quality you will feel for years. Put nowhere you can see, it buys margin you are paying for.
What should you not expect AI to fix?
A faster way to produce code does not fix the things that make software projects go wrong. It will not rescue a vague brief, because it will build the wrong thing more quickly. It will not replace the people who understand your business well enough to decide what should be built in the first place. And code generated without a careful eye can cost you more later, when someone has to understand and maintain it.
The most expensive mistake in any build is not slow typing. It is building the wrong thing, and that is settled in discovery and planning long before a line of code is written. The faster coding tools sit underneath that work, not in place of it, and any team worth hiring treats them the same way.
Frequently asked questions
Can I just use AI to build my app myself?
You can build a rough prototype, and that is genuinely useful for testing an idea before you spend on it. Turning that prototype into something a business can rely on, with real data, proper security, and other systems connected, is where the harder and less automatable work begins. A common and sensible path is to sketch the idea with AI, then bring in a team to build the version your customers actually use.
Does AI written code mean lower quality?
Not by itself. Generated code is only as good as the review around it. With a developer checking, testing, and correcting it, quality holds up fine. Without that check, it tends to add bugs and maintenance cost. The useful thing to ask any team is who reviews the generated code and how it gets tested.
Will my project cost less because of AI?
It can, in proportion to how much of your build is predictable, repetitive work, and less so when discovery, integration, and judgement dominate. The honest way to know is to scope your specific project rather than apply a general percentage you read somewhere.
Who owns and maintains AI assisted code?
You should own all of it, exactly as you would hand written code, with the full source and documentation handed to you. Maintenance works the same way too. Code has to be understood by a person before it can be changed safely, which is why how it was written and documented matters far more than whether a tool helped produce it.
Scope your build with the project estimator
AI has made some of the work of building software faster, and more capable models keep arriving. Whether that reaches your quote depends entirely on what your project is made of. The way to find out is to scope the actual work rather than trust a headline percentage. Start with our project estimator to see the shape of your build, then talk to us about where AI genuinely saves time on it and where careful human work still earns its place.
- AI clearly speeds up the predictable parts of a build, boilerplate, routine tests, migrations and first drafts of standard screens, where the right answer is largely known in advance
- It barely helps with discovery, architecture, judgement and testing with real users, and on large familiar codebases it can even slow experienced developers down
- Whether a faster build reaches your quote depends on how much of your project is the predictable kind of work, and on whether the agency passes the saving on
- Ask one question of any agency: where does AI save you time on my build, and how does that show up in what I pay