Here is something you will rarely hear from a developer who wants your business: sometimes you do not need one. Sometimes an AI tool is genuinely enough, and the honest answer is "use the AI, save your money, come back when you actually need me." I would rather tell you that up front than sell you a £10,000 rebuild you did not require.
But the opposite is also true, and it is the part that quietly costs people the most. There is a clear line where AI tools stop being a clever shortcut and start being a liability — and the trouble is, that line is invisible from the demo. Everything looks finished. It works on your laptop. Then it goes live, real customers turn up, and it falls over in ways you did not know to worry about. This article is about finding that line before it finds you.
Why this question even comes up now
A couple of years ago, "build it yourself" meant WordPress and a weekend of frustration. Today you describe what you want to Lovable, Bolt, v0 or Cursor, and forty seconds later you are looking at a working app. Vibe coding — Collins Dictionary's word of the year for 2025, so we can drop the term without explaining it — has genuinely changed who can ship something.
That is brilliant. It is also where the confusion starts. Because the gap between "looks finished" and "is actually safe to run a business on" has never been wider. AI gets you about 80% of the way with almost no effort, and that last 20% — the unglamorous part with the payments, the data protection and the edge cases — is exactly the part it cannot reliably do. So the real question is not "can AI build it?" It usually can. The question is "is this one of the things where the missing 20% will hurt me?"
Where AI genuinely is enough
Let me be specific, because vague reassurance helps nobody. There is a whole category of work where an AI tool is not a compromise — it is the correct, sensible, money-saving choice, and bringing in a developer would just be burning cash.
This is where an AI tool is more than enough
- A hobby or personal landing page. A one-pager for your side project, your CV, your wedding, your podcast. Nobody is paying you through it and no customer data is at stake. Lovable or a free Vercel deploy is perfect.
- An internal demo or pitch. You need to show an investor or your team what the idea feels like. It only needs to survive a ten-minute walkthrough on your screen, not a thousand real users.
- Visualising an idea before you commit. You are not sure the concept works. Spinning up a clickable prototype with v0 or Bolt to test it on a few people is far smarter than paying anyone to build the real thing first.
- A static brochure site with no logic. A few pages of text and images, a phone number, maybe a link to your email. No login, no payments, no database. AI plus a static host does this cleanly.
- Throwaway internal tools. A quick calculator, a rota, a single-purpose form your team uses behind a password. If it breaking just means a mild annoyance for you, AI is fine.
If your project lives in that box, my honest advice is: stop reading sales pages, open the tool, and ship it. You do not need me, and I will happily tell you so. Spending senior developer money here would be like hiring an architect to put up a garden shed.
Where AI runs out of road and you need a person
Now the other side. These are the things where the demo will look just as polished, but the moment it touches the real world it carries risk that an AI tool simply does not know exists. This is where you need a human who can be held responsible for what gets shipped.
This is where you need a developer who owns the outcome
- Real payments. The moment you take money — Stripe, PayPal, GoCardless Direct Debit, Apple Pay, Klarna — you are in PCI-DSS territory and you cannot afford a "works on the demo" checkout. Hosted Stripe or PayPal flows keep you out of the worst of it, but they have to be wired up correctly, with proper webhooks, error handling and refunds. AI tools routinely leave secret keys exposed or skip the failure paths entirely.
- Real customer data. If you store names, emails, addresses or anything personal, UK GDPR and the Data Protection Act 2018 apply to you — there is no small-business exemption. You need a lawful basis, a privacy policy, PECR-compliant cookie consent and an ICO registration. AI-generated apps almost never ship a single one of these.
- Scale. A demo serving you is not the same as an app serving two thousand people at once. Database queries that are fine for one user grind to a halt under load. The AI never tested that, because it only ever saw one user: you.
- Integrations. Connecting to Xero or QuickBooks for invoicing, Royal Mail or DPD for shipping, a CRM, an SMS gateway, HMRC Making Tax Digital — these are fiddly, real-world contracts that AI tools fudge or fake. I wrote a whole separate piece on exactly why AI struggles to connect real payment and business integrations.
- Security. Insecure Supabase or Firebase rules are the single most common thing I find. The AI leaves the database wide open so anyone who knows the URL can read every record. It works perfectly in the demo precisely because nobody is attacking it.
- Maintenance and ownership. Production software is never "done." Things break, dependencies update, browsers change. You need someone who can read the code, fix it, and keep it running — not a chat window that confidently rewrites three things and breaks two others.
The common thread is accountability. When real money, real data or other people depending on you is on the line, you need someone who can say "I have checked this, it is secure, and if it breaks I will fix it." An AI tool cannot make that promise. It does not even know it should.
AI versus a developer: where the line actually sits
Here is the same decision laid out side by side. Find the row that matches your project and you will usually have your answer.
| What you are building | AI tool is enough | You need a developer |
|---|---|---|
| Stakes if it breaks | Mild annoyance, you fix it later | Lost sales, leaked data, legal exposure |
| Money flowing through it | None | Real payments, subscriptions, refunds |
| Personal data stored | None, or only your own | Customer data — UK GDPR applies |
| Number of users | You and a handful of people | Hundreds or thousands, concurrent |
| External integrations | None | Stripe, Xero, Royal Mail, CRM, HMRC |
| Lifespan | Days or weeks, then thrown away | Years, needs maintenance |
| Honest cost | Free to ~£40/month + your time | £3,000–£15,000+ to do properly |
The decision checklist: AI or a person?
If you want it simpler than a table, ask yourself three questions in order. Stop at the first "yes."
Three questions, in order
- Will real money move through this app? If yes — you need a developer. Payments are not a place to learn on the job.
- Will it store other people's personal data? If yes — you need a developer. UK GDPR and the ICO do not care that an AI wrote it; the liability is yours.
- Will other people or businesses depend on it staying up? If yes — you need a developer. Uptime, scale and maintenance are human responsibilities.
Three honest "no"s? Use the AI tool with my blessing. Even one "yes" and you are in territory where the cheap path quietly becomes the expensive one.
How the checklist plays out in practice
A photographer in Manchester wants a portfolio site to show her work and a contact email. No payments, no logins, no stored data. Three "no"s. She builds it in Lovable in an afternoon and it is genuinely the right call.
A founder in Shoreditch vibe-codes a booking app for a niche service. It looks fantastic. But customers pay through it (yes), it stores their names and card details flow through it (yes), and salons rely on it being up every morning (yes). Three "yes"s. The AI build was a brilliant way to validate the idea cheaply — and exactly the wrong thing to run a real business on without a proper production rebuild.
Same tool, same forty seconds of generation, completely different correct decision. That is the whole point: it is never about whether AI can build it. It is about whether the missing 20% will hurt you.
What each path actually costs
Let me put real numbers on both routes so you can compare like for like. UK quotes are ex-VAT by convention, so I will say "+ VAT" where it applies.
The AI path (you do it yourself)
The tool itself is cheap. Lovable, Bolt, v0 and Cursor sit somewhere between free and roughly £20–£40 per month (~€23–€47). A static host like Vercel or Netlify is free for small projects. For a hobby landing page or an internal demo, your only real cost is your own time — and that is entirely fair, because for that kind of work this is the correct, frugal choice.
The cost that catches people out is not the subscription. It is the day a free AI site grows a payment form, a login or a contact form that quietly starts storing data — because that is the day you have, without noticing, taken on PCI scope, GDPR obligations and a security surface you cannot see. I unpicked exactly how a "free" AI site stops being free in this companion piece on taking a ChatGPT idea to a real app.
The developer path (I take it over)
When real risk is involved, here is what senior independent UK work actually costs in 2026. These are the numbers I would quote, and they are the numbers that mean someone can genuinely ship to production rather than a £5-an-hour marketplace gamble.
| Engagement | Typical price (ex-VAT) | What you get |
|---|---|---|
| AI app health check / audit | £500–£1,500 (~€590–€1,750) | A plain-English report on what is unsafe, what is missing and what it would take to fix |
| Rescue / finish my AI-built app | £3,000–£15,000+ | Secure the data, wire payments properly, GDPR done, production-ready |
| Senior developer day rate | £400–£650/day | Someone who can actually ship to production, not just to a demo |
| Landing / one-pager (built properly) | £800–£2,500 | For when even a brochure site needs to be fast, accessible and reliable |
| Custom MVP / web app | £8,000–£30,000+ | A real product, built to grow, with payments and data done right |
Notice the audit at the bottom of that price range. It exists precisely so you do not have to guess. For a few hundred pounds you find out whether your AI build is fine as-is, needs a tidy-up, or needs a proper rebuild — before you spend the bigger money.
The common mistake: asking AI to do the dangerous bit
The pattern I see most often is not someone using AI for the wrong thing entirely. It is someone using it brilliantly for the right 80% — the screens, the layout, the flow — and then assuming it handled the dangerous 20% too. It did not. It generated something that looks like a working payment flow and a secure login, because looking right is what these tools are best at. Under the surface, the database is open, the keys are exposed, the consent banner is decorative and there is no error handling at all.
This is not a reason to avoid AI. It is a reason to know which 20% you are standing on. Validate cheaply with the tool, absolutely — then bring in a person for the part where "it worked on the demo" is not good enough.
My honest position
I make my living rebuilding and rescuing apps, so you might expect me to tell you that you always need a developer. I will not, because it is not true. If your project is a hobby page, an internal demo or an idea you are still testing, use the AI tool, keep your money, and move fast. I genuinely mean that — and if you are not sure which side of the line you are on, that is exactly the conversation I am happy to have for free.
But when real money, real customer data or real dependence on uptime enters the picture, "the AI built it and it looked fine" is the most expensive sentence in software. That is the moment to bring in someone who audits it, secures it, makes it compliant, and owns the outcome when it goes live. Knowing which moment you are in is the whole skill — and now you do.
Not sure which side of the line your project falls on?
Send me a short description of what you are building or what you have already vibe-coded. I will tell you honestly whether AI is enough — or whether it needs a proper, secure rebuild. No hard sell.
Get an honest readFrequently asked questions
How do I know whether an AI tool is enough or I need a developer?
Ask one question: does the app handle real money, real customer data, or other businesses depending on it staying up? If the answer is no, an AI tool is usually enough. If the answer is yes, you need a developer who can take responsibility for security, GDPR compliance and uptime in production.
Can AI tools replace a developer entirely?
For a static landing page, an internal demo or visualising an idea, yes. For anything touching payments, personal data, integrations or scale, no. AI gets you roughly 80% of the way and the last 20%, where the real risk lives, still needs a person who understands what they are shipping.
How much does it cost to build a website myself with AI?
The tool subscription is typically free to around £20–£40 per month. A hobby landing page can genuinely cost you nothing but your own time. The hidden cost appears the moment you add a payment form, a login or a contact form that stores data, because that is where compliance and security work starts.
How much does it cost when a developer takes the project over?
A fixed-price health check or audit of an AI-built app is typically £500–£1,500 + VAT. A production rebuild or rescue usually lands between £3,000 and £15,000 + VAT depending on scope. Independent senior UK day rates run £400–£650 + VAT, and a new custom MVP is usually £8,000–£30,000 + VAT.
I started with AI and got stuck. Do I have to throw it all away?
Almost never. The screens, the copy and the product thinking you validated are real progress and worth keeping. What usually gets rebuilt is the layer underneath: the database rules, authentication, payment flow and data handling. The visible part you already have is a head start, not wasted money.
Should I just wait until AI is good enough for everything?
No. AI is genuinely good at the parts it is good at today, and you can ship those parts now. The parts that need judgement, accountability and compliance are not a waiting game; they are a different kind of work. Use AI where it already wins and bring in a developer for the part that carries real risk.
Want the bigger picture? Start with my homepage and app development services, then read the companion pieces on why AI won't connect real integrations and how to take a ChatGPT idea all the way to a real app.