What AI Can't Do: Why You Still Need an Accountable Human

Hello! Mike here. If you have spent any time on LinkedIn or YouTube lately, you have seen the message: anyone can build a website with AI now. No designers, no developers, no agencies. Type a prompt, get a site, done.

So why would you pay an agency?

Fair question. And before I answer it, let me be clear about who is answering. I run Glide, an AI-native agency. AI agents publish content for me overnight. AI writes most of our code. I have written publicly about how I run the whole agency on AI, with commit logs as proof. I am about the furthest thing from a web designer defending his old job from the robots.

Which is exactly why you should take this seriously: even I will tell you there are things AI cannot do. Not "cannot do yet". Cannot do, structurally, because they are not building problems. They are accountability problems.

First, the honest bit: yes, AI can build you a website now

I am not going to strawman the tools, because they are genuinely good. The current AI builders can produce a clean, modern, mobile-friendly site in an afternoon. The copy will be passable. The design will be fine. For a hobby project or a simple brochure site with nothing riding on it, that might honestly be all you need.

The whole industry knows this. Webflow is restructuring itself around agentic tools. Wix laid off around a thousand people while rebuilding the company around AI. Every platform you have heard of is racing to make "type a sentence, get a website" real, and they are largely succeeding.

Here is what that shift actually leaves behind for agencies like mine: the work nobody wants. The migration with no clean export path. The thing that breaks at 2am. The half-finished AI project that needs someone to make it actually work. I know because that is increasingly what lands in my inbox.

So no, the pitch is not "AI sites are rubbish". The pitch is six specific things the tools cannot do, each with a real-world consequence. Here they are.

1. Be accountable when it breaks

Every website breaks eventually. An update goes wrong, a plugin conflicts, a host migrates something badly, a renewal lapses. The question is never if. It is who picks up the phone when it does.

Last week's example from our own client list. A skincare ecommerce store had its entire site wiped by a botched core update. Not "a bit broken". Gone, with the database overwritten by demo content. Orders, products, years of work. For an online store, that is the business switched off.

We restored it from backup and had it live again in a day. The client left an unprompted five-star review, and the line that stuck with me was the relief in it. Not "great design". Relief.

Now run the same scenario on a DIY AI build. Who do you call? The builder's support queue? A help forum? The AI that made the site has no memory of you, no backup plan it owns, and no skin in the game. Nobody is paid to care that your store is down. That gap costs nothing right up until the day it costs you everything.

2. Know your business

An AI builder knows what you typed into the prompt box. That is it. It does not know that half your enquiries come from one referral partner, that your margins are better on one service than another, or that the customers you actually want keep mistaking you for a cheaper competitor.

That stuff comes out in conversation. Every good website decision I have made for a client started with something they said on a call, usually something they did not think was important. The page structure, the offer, what goes above the fold: that is strategy, and strategy comes from a human asking questions, not from a prompt.

The consequence of skipping it is subtle, which is what makes it dangerous. You get a site that looks finished and says nothing. It describes your industry instead of your business. Visitors read it, feel nothing, and click back to Google.

3. Say no to you

This one matters more than people realise. AI tools are agreeable by design. Ask one to build a website for a terrible idea and it will build it enthusiastically, beautifully, and fast.

Part of what you pay a decent agency for is the no. No, do not put all nine services on the homepage. No, that price positioning will attract the wrong clients. No, you do not need a blog you will never write. I have talked people out of work they were ready to pay me for, because it would not have got them anywhere, and a client who wins stays for years.

An AI yes-machine will never do that. It will build the wrong thing to a very high standard, and you will not find out it was the wrong thing for six months.

4. Judge quality

Here is the uncomfortable secret of AI production: AI generates quickly but judges poorly. It cannot reliably tell its own good output from its own slop, because everything it makes looks plausible to it.

I see this daily, because AI drafts most of what Glide produces. The drafts are good. They are also wrong in small, confident ways: a claim that is not quite true, a paragraph that sounds like every other website on the internet, a layout choice that technically works and feels off. Catching that takes taste, and taste is still entirely human.

Most AI-built sites never get that filter. That is why so many of them share the same hero section, the same three-column feature grid, the same vaguely motivational copy. The consequence is a website that quietly signals "nobody senior looked at this", and customers pick up on it even when they cannot say why.

5. Handle the messy middle

The tutorials always show the fun part: prompt in, pretty homepage out. They never show the messy middle, because the messy middle does not demo well.

Migrating ten years of content off a dying platform without losing your Google rankings. Pointing DNS without taking your email down for two days (this one bites more businesses than anything else). Connecting the booking system, the payment provider, the CRM, the newsletter tool, and making them all agree about who your customers are. Redirect maps. SSL quirks. The contact form that silently stopped sending a month ago.

None of that is glamorous, and all of it is the actual job. It is precisely the work AI platforms leave with you, and it is where DIY projects stall, half-live, with email broken and the old site still ranking. We have built sites in days rather than months using AI, and I can tell you the build is now the quick bit. The messy middle is what you are really hiring for.

6. Care about the result in 12 months

A website is not a launch. It is the start of a feedback loop. Somebody has to watch Search Console, notice which pages Google is rewarding, fix the ones it is ignoring, and act on what the data says, month after month. That ongoing technical work is a discipline in itself, and I have written up exactly how we approach it if you want the detail.

An AI builder does not do this. Not because it technically could not, but because nobody is responsible for it. The tool's job ended at publish. Most owners open Search Console once, feel mildly confused, and never go back. Twelve months later the site that looked great at launch is invisible, and nobody noticed it happening.

The sites that win are not the ones that launched best. They are the ones somebody kept caring about.

The real question is not AI or human

So here is where I land, as somebody who uses these tools more than almost anyone you will meet.

"Should I use AI to build my website?" is the wrong question. AI will be involved in building your website either way. If you do it yourself, AI builds it. If you hire us, AI builds most of it too, and I say that openly. The right question is: who is accountable for the outcome?

With a DIY AI build, the answer is you. Your evenings, your DNS records, your 2am problem, your Search Console nobody checks. For some businesses that is a fine trade. Genuinely.

Glide's answer is different: AI does the production, and a named human answers for the result. Me. Not a ticket queue, not a chatbot, a person you can call whose reputation is attached to your website working. We use whatever platform fits the job, WordPress, Shopify, Next.js or fully custom, because we are not married to any of them. The tools are interchangeable. The accountability is the product.

What that actually costs

Here is the part that surprises people: this model is cheaper than the old agency model, not more expensive, and AI is the reason why.

The old model charged you thousands up front because human production time was the cost. AI removed most of that cost, so we removed the big invoice. Glide websites are a monthly subscription, starting at £150 per month, with most clients on £350 per month including the ongoing work I have spent this whole post arguing matters most: the maintenance, the watching, the judgement, the human who picks up the phone.

You are no longer paying anyone for typing speed. You are paying for taste, strategy and accountability. Those got more valuable as the typing got free, not less.

If you are weighing up an AI builder against hiring someone, I am happy to give you a straight answer either way. Sometimes the honest answer is "use the builder, you do not need us yet", and I have said exactly that on calls before. Book a chat and ask me.

Mike McDonnell, Founder of Glide Marketing. More about how I work.