Will AI build my website? What actually changes for a small business in 2026

Every few weeks someone asks me a version of the same question. Should I wait? If AI is getting this good this fast, why pay someone to build a website now, when I might be able to describe one into existence for free next year?

It's not an unreasonable thing to wonder. AI use among UK businesses overall reached 25% by December 2025, up 15 percentage points on late 2023, according to the ONS. Among the smallest firms specifically, those with fewer than ten employees, the figure sits at 28%. The pace is real. The question is what that pace actually changes for someone commissioning a website, not building one.

It's a fair question. It deserves a straight answer rather than the two lazy ones on offer. One lazy answer is "AI is a fad, hire a professional." That isn't true, and anyone saying it either hasn't used the tools or has a reason to pretend they haven't. The other lazy answer is "AI does the work now, so you don't need a professional." That isn't true either, and it's usually being sold by someone with something to gain from you believing it.

I use AI tools daily. Every project I ship runs through them somewhere. So here's the honest version, from someone with nothing to lose by telling you straight: no, you shouldn't wait, but not for the reason either lazy answer gives you. AI has changed how fast a website gets built. It hasn't changed who's responsible for deciding what to build.

---

What AI actually changes

AI has genuinely compressed the mechanical parts of building a website. Scaffolding a page structure, generating a first pass at layout options, writing boilerplate code, checking a design against accessibility standards: work that used to take hours now takes minutes. Research from McKinsey backs this up with a number rather than a feeling: AI cuts time spent on routine coding tasks, boilerplate, tests, documentation, by around 46%. The same research is explicit about what's excluded from that figure: system design, debugging complex issues, and understanding an unfamiliar codebase. Those are the tasks that don't get faster, because they're the ones that were never mechanical in the first place.

This isn't a one-sided story, and I'd rather tell you the honest version than the flattering one. A 2025 randomised controlled trial by METR gave experienced developers AI tools and measured the result directly: task completion took 19% longer, even though those same developers believed the tools had made them 20% faster. The gap between how fast AI feels and how fast it actually is deserves more attention than the hype gives it. Judgement about when a tool genuinely helps, and when it just feels like it's helping, is exactly the kind of decision that still needs a person paying attention.

That's real, and it's the reason a well-run studio can now deliver in weeks what used to take months. The cost of a UK small business website has come down over the last few years, and this is a large part of why. Faster iteration means more design options tested per project, not fewer. It means a small studio can offer the kind of bespoke attention that used to be reserved for clients with much bigger budgets, because the mechanical overhead that ate into that attention has shrunk.

Here's the part worth sitting with: the value doesn't disappear when the timeline shrinks. It moves. AI compresses build time. It does not compress thinking time. Working out what your site actually needs to do, which of your customers it needs to convert, and which features are worth building at all: that's judgement, and judgement was never a speed problem. It's the part AI can't shortcut, because it depends on knowing your business, your customers, and what you're actually trying to achieve. A hand-curated brief still produces a better result than a generic prompt, however capable the model behind it gets, for the same reason a brief conversation with someone who knows your market beats a form filled in alone.

Put plainly: faster build, the same amount of thinking required, more bespoke work possible per pound than five years ago. That's the honest shift. It isn't smaller than the hype suggests, and it isn't the shift the hype implies either. It's a change in where the value sits, not a removal of the value.

What AI doesn't change

Three things AI hasn't touched, and I don't expect it to any time soon.

Changes with AI	Doesn't change
Layout and code scaffolding speed	What your site needs to do for your customers
Number of design directions explored per project	Knowing which features to leave out
Time spent on routine, repeatable build tasks	Matching the brand, not the median AI output
Overall project cost and timeline	Who answers when something breaks at 9pm

Knowing what not to build. The tools will happily generate ten more pages, three more animations, a chatbot you don't strictly need. They don't push back, because they have no stake in whether the result actually serves your customers. A good brief usually has almost as much removed from it during scoping as added to it, and that removal is entirely a judgement call: what serves the customer, what serves someone's ego, and the difference between the two. That's a conversation, not a prompt.

Brand fit. AI-generated output defaults toward the generic, because it's trained on the average of everything that came before it. Left unsteered, it converges toward every other AI-assisted site: the same layout patterns, the same safe phrasing, the same "good enough" choices that don't offend anyone and don't stand for anything either. Getting something that looks and reads like your business, rather than the median output of a model, takes a person checking every result against what actually fits your brand, not what's statistically probable across every business that used the same tool.

Trust. If something breaks at 9pm the night before a launch, or a client needs a straight answer about why a decision was made, that's a person picking up the phone. No amount of AI assistance changes who answers when it matters. Ownership of that relationship, and accountability for what gets shipped, still sits with a person.

What this actually looks like in practice

I recently rebuilt the site for Orchard Classic Decals, a Kawasaki-focused vinyl decal business that had been running on a twelve-year-old Shopify theme. The brief conversation was the usual one: what's actually stopping browsers from buying, which pages carry the eBay reputation (891 lifetime reviews, effectively zero negative feedback) that the old site did nothing to surface, what needed to stay identical because customers already knew how to use it. Entirely human, and it should be, because getting that wrong wastes the whole project.

Once the brief was locked, the build moved fast: live in 26 active days, from first commit to launch. Layout options that used to take days to mock up got explored in an afternoon, so decisions came from seeing three genuinely different directions rather than one safe option with minor variations. Routine code, the parts of every project that are the same regardless of client, got written faster, which freed up time for the parts that were actually specific to this business: the copy, the model-line browse structure, the small decisions about hierarchy that separate a site that works from one that merely exists.

The checkpoints didn't change. The homepage got reviewed before anything else was built. Feedback got incorporated by someone who understood why a particular change was requested, not just what was typed into a box. A Google Lighthouse performance check and a proper look at what Google Search Console was flagging happened before launch, same as they always would. Launch was the start of a working relationship, not the end of one.

What's different is the ratio. More of the paid time goes toward decisions that need a person, because less of it is being spent on mechanical setup. That's the honest version of "AI-assisted": not a shortcut around expertise, but a shift in where the expertise gets spent.

So should you wait?

No. Not because "wait" is an inconvenient word for me to hear, but because the maths doesn't favour it. AI tools are useful today, in the hands of someone who knows how to steer them well. Waiting for them to become good enough to remove the person entirely means waiting for a shift that keeps not fully arriving, while your current site keeps costing you the customers who land on it and leave before they convert. A professionally built UK small business site typically runs £500 to £8,000 depending on scope, and that band has already come down as AI-assisted build times have shortened. Waiting for it to drop further on the promise of full automation is betting against a trend that keeps flattening out short of the finish line.

There's also a version of "waiting" that's really avoidance dressed up as strategy. If the honest reason you're hesitating is budget, or genuine uncertainty about what your site actually needs to do, that's a different and far more useful conversation than "should I wait for the technology to catch up." One of those questions has a clear answer. The other deserves a proper diagnostic before anyone quotes you anything.

What to actually do instead

If you're weighing a website project now, three things matter more than the AI question.

Get clear on the outcome you want. More enquiries, more bookings, a site that finally matches how good the business actually is. The mechanism behind the build, whatever tools sit inside it, is secondary to that outcome. Start there and the rest of the decision gets easier.

Find someone who'll show you the working, not just the output. Ask how they use AI in their process, and listen for specifics rather than a confident shrug. A straight, detailed answer tells you more about someone's judgement and their honesty than any portfolio ever will.

Don't confuse cheaper with faster. A compressed timeline built on genuine expertise and a clear brief is a good trade for everyone. A compressed timeline built on skipping the thinking is a site you'll be quietly replacing again within the year, and paying for twice.

AI is the hammer. The person swinging it still decides what gets built, and why it's worth building at all. That hasn't changed. Honestly, I don't think it's going to.

References

• Whito (2026). UK Business AI Adoption Statistics 2026, citing ONS Business Insights and Conditions Survey (December 2025 fieldwork). https://whito.co.uk/research/uk-ai-adoption-statistics/
• Second Talent (2026). AI Developer Productivity: 2026 Data, citing McKinsey & Company research on routine coding task time reduction. https://www.secondtalent.com/resources/ai-developer-productivity-tools-2026/
• Index.dev (2026). Top 100 Developer Productivity Statistics with AI Tools 2026, citing METR randomised controlled trial (2025). https://www.index.dev/blog/developer-productivity-statistics-with-ai-tools