You Can't Build a Business in 20 Minutes

Contra the Hype · Opinion · Technology & Code · April 2026

There is a particular kind of content doing the rounds at the moment. A confident voice, a screen recording, an AI agent being prompted to reproduce some popular SaaS tool, and then — within minutes — a breathless claim that the entire software industry has just been disrupted. The indie developer is finished. The senior engineer is obsolete. Why would anyone pay a subscription when you can just build the whole thing yourself in an afternoon?

I want to push back on this, firmly and with specifics. Not because I'm threatened by AI coding tools — I use them daily and they are genuinely transformative — but because the narrative being sold is not just wrong, it's misleading in ways that will cost real people real money and real time.

The Demo Is Not the Product

Yes, you can prompt Claude Code, Cursor, or any capable agent to scaffold something that visually resembles a known application. It will generate routes, render a UI, wire up some state. It will look, in a demo, remarkably like the real thing. But a demo is not a product, and a product is not a business.

What the demo doesn't show you is the next 200 hours. The authentication edge cases. The database migrations that break on the third deploy. The email delivery that silently fails in production. The rate limiting you forgot to implement. The input validation that leaves you wide open. The session handling that works perfectly until it doesn't, at 2am, for your only paying customer.

Real software deployments — even modest ones, even well-scoped ones built by experienced teams using the best AI tooling available — take hundreds of hours. Not because developers are slow or inefficient, but because software of any real value is genuinely complex. There are hundreds of decisions embedded in any working system. What happens when the payment fails halfway through? What happens when two users edit the same record simultaneously? What happens when your file upload receives something unexpected? These aren't edge cases you can handwave. They're the product.

AI makes mistakes. Confidently, fluently, and at scale. Without the knowledge to recognise them, you won't know what you've built — only what it looks like.

You Still Have to Understand What You Built

Here is the thing that the AI hype content never addresses: if you don't understand the code, you don't know what's in it.

This isn't gatekeeping. This is a practical reality with serious consequences. An AI agent will generate code that is, on the surface, entirely plausible. It will follow conventions. It will pass a casual read. But it may also contain subtle logical errors, insecure patterns, missing validations, or architectural choices that will make the whole thing impossible to extend in six months. And if you can't read the code, you cannot see any of that.

You don't know if your user passwords are being hashed properly. You don't know if your API endpoints are authenticated. You don't know if the SQL being built from user input is safe. You don't know if the data you're storing complies with GDPR. These aren't hypothetical concerns for a security researcher — they are the baseline requirements for running anything that touches real users and real data. An AI can introduce any of these issues silently, and without the ability to review what's been written, you have no way of knowing.

Copying Software You Don't Want to Pay For

A specific variant of this fantasy deserves its own treatment: the idea that instead of paying £5 a month for a tool you like, you can just have an AI reproduce it for you and run it yourself for free. This argument collapses almost immediately once you apply any scrutiny.

Firstly, you still have to run what you've built. Software doesn't run in a vacuum. You need a server, or a hosting provider, or a cloud platform. You need a domain. You need SSL certificates. You need a database with backups. You need to handle deployments when something breaks. Even a modest self-hosted setup has an ongoing operational cost — in money, and more significantly, in time and expertise.

Secondly, what you've replicated isn't the product — it's the interface. The tool you're paying £5 a month for has years of reliability engineering, customer support, uptime monitoring, and iteration behind it. What AI generates in twenty minutes has none of that. The moment something breaks on your self-hosted clone — and it will — you own the problem entirely. There's no support ticket. There's no status page. There's just you, a codebase you don't fully understand, and an outage.

Thirdly, the economics rarely work the way people imagine. Once you account for compute costs, developer time, maintenance overhead, and the opportunity cost of not working on something more valuable, self-hosting a clone of a £5/month tool is almost never cheaper. It's just the illusion of free because the costs are less visible.

What AI Actually Does to the Software Industry

Now here is where I want to be genuinely optimistic, because I think the real story is more interesting and more valuable than either the hype or the doom.
AI coding tools are not killing developers. They are dramatically expanding the leverage of developers who know what they're doing.

A capable engineer who previously needed six months to ship a product can now ship it in six weeks. Not because the hard parts have vanished, but because the tedious scaffolding, the repetitive boilerplate, the routine CRUD code — all of that accelerates enormously. What remains is the judgment work: architecture, security decisions, edge case handling, user experience intuition, business logic. The things that require genuine understanding and genuine experience. The things AI cannot substitute for, only assist with.

This is the actual disruption. The floor for what a solo developer or a small team can build has dropped dramatically. The ceiling for what an experienced developer can ship has risen just as dramatically. That is a genuinely exciting shift. It means more software gets built, and more of it gets built by people who care deeply about what they're making.

AI hasn't eliminated the need for expertise. It's made expertise more productive than it has ever been.

What it does not mean is that expertise is optional. The developer who understands their stack, who can read a generated diff and spot the problem, who knows which shortcuts are safe and which will hurt them in three months — that developer gets enormously more powerful with these tools. The person who cannot evaluate what's been generated gets an extremely convincing-looking liability.

Business Intellect Is Not Optional Either

There is a related myth worth dismantling: that the barrier to running a software business was always just the code, and now that AI can write the code, anyone can run one. This fundamentally misunderstands what running a software business involves.

You still need to understand your users. You need to price correctly, market effectively, handle support, manage infrastructure costs, deal with billing edge cases, navigate legal compliance, and make the thousand small decisions that determine whether a product thrives or quietly dies. None of this is in the repository.

The people who will succeed with AI coding tools are those who combine technical literacy with business judgment. They don't need to be 10x engineers or MBAs — but they do need to understand the craft well enough to steer it, and the business well enough to know what they're building toward. The tools amplify these qualities; they cannot replace them.

Why This Matters

I am not making this argument out of professional defensiveness. I genuinely believe AI tooling is one of the most exciting developments in software in my lifetime, and I use it as a core part of how I work. But I've also watched people burn significant time and money on the promise of the 20-minute build, only to find themselves with something half-finished, insecure, and impossible to maintain.

The hype is not neutral. It sends people down paths they aren't equipped for. It devalues the real expertise of people who spent years learning their craft. And it sets unrealistic expectations that cause genuine harm when they inevitably collide with reality.

The truth is more nuanced, and ultimately more hopeful: AI is making great developers faster, not making development unnecessary. The developers who understand what they're building — who can direct these tools with judgment and verify their output with knowledge — are going to ship things that would have been impossible for a small team just a few years ago. That's genuinely worth celebrating.

But you can't outsource the understanding. You never could.

The indie developer isn't dead. In many ways, this is their best moment yet — if they do the work to actually know what they're building. Weeks instead of years is real. Twenty minutes instead of years is a video.

Know the difference.