The AI Gold Rush Won't Wait for You to Get Your Shit Together

I think we’re living through one of the best opportunities in history to start a software company, and most people are going to miss it because they’re too busy arguing about whether AI-generated code counts as real programming.

I’m not talking about launching another chatbot wrapper that charges $19 a month to summarise PDFs. The world has enough of those. I’m talking about one person sitting at home, building and shipping something that would’ve required a funded startup and a small engineering team a few years ago.

And doing it in a weekend.

That should sound completely fucking ridiculous. Five years ago it would’ve been.

Today? I’m seeing it happen. I’m doing it myself.

The interesting part isn’t that AI makes software development faster. It’s that the cost of trying something has collapsed. And there’s a window, probably a fairly small one, between that happening and everyone else figuring out what it means.

Most people haven’t even paid for this stuff yet

Here’s a statistic that should make anyone with an entrepreneurial bone in their body sit up.

According to a16z’s October 2026 consumer AI report, just 4.5% of eligible consumers in a US spending panel had an active paid personal subscription to ChatGPT, Gemini or Claude in August. That was up from 2.1% a year earlier.

That’s a measure of subscriptions to three major assistants, not every AI product on the planet, and the panel isn’t a census. Other surveys get much higher numbers when they ask people whether they pay for any AI tool, including subscriptions paid for by employers. Different questions, different answers.

But the point stands: paying for a mainstream AI assistant is nowhere near universal.

The same a16z analysis found that only 13% of people paying for one AI product also paid for another. And the top 1% of AI spenders averaged $903 a month on consumer cards.

Nine hundred and three dollars. Every month.

While a huge chunk of the population is still asking ChatGPT to write a birthday message for their aunt, a small group is spending serious money using these tools to build things.

That doesn’t mean everyone else is stupid. Most people have jobs, families and absolutely no reason to care about the difference between a coding agent and an autocomplete plugin.

It does mean there’s a pretty enormous gap between what this technology can do and what most people currently understand it can do.

And gaps like that don’t tend to stay open forever.

The barrier to building software has fallen through the floor

I’ve been writing software professionally for over 18 years. I remember when launching a relatively ordinary web application meant weeks of scaffolding, configuring servers, wiring up authentication, writing boilerplate, setting up deployments and eventually remembering you still had to build the actual product.

Then you’d find a bug in production at 11pm and spend three hours discovering you’d misspelled a database column.

Character building stuff.

Today I can describe a feature to a coding agent, have it inspect an existing codebase, implement the changes, write tests and give me something to review while I work on something else.

It’s not magic. It gets things wrong. Sometimes spectacularly wrong. If you blindly ship whatever comes out, you’re eventually going to discover why security engineers drink.

But the amount of work one capable person can get through has changed dramatically.

And it’s not just code.

You can use AI to help research a market, analyse customer feedback, draft documentation, produce marketing material, explore designs, write support responses and automate tedious internal processes.

A solo founder can now cover an absurd amount of ground without hiring six people before earning their first dollar.

That doesn’t mean AI replaces experience. If anything, experience is a multiplier. Knowing what good software looks like, where systems fail, how to evaluate generated code and what customers actually need is still enormously valuable.

But the price of getting from an idea to a working product has dropped so far that the old excuse of needing a team and a few hundred thousand dollars is starting to sound a little dated.

For some products, you need a laptop, a few subscriptions and the willingness to look like an idiot until something works.

That last part has always been free.

The real opportunity isn’t building another AI app

Every time there’s a new technology boom, people rush to build products about the technology.

AI chatbots. AI note takers. AI email writers. AI meeting summaries. AI tools that turn a sentence into a slightly longer sentence and charge you monthly for the privilege.

I’m not saying none of those are useful. Some are excellent businesses.

But I think the more interesting opportunity is using AI to build ordinary, useful software for problems that have been ignored because they weren’t worth the development cost.

Think about the local business still running its scheduling through spreadsheets and text messages. The small logistics company juggling three disconnected systems. The niche industry paying a fortune for software that looks like it was last updated when Internet Explorer was considered cutting edge.

There are thousands of problems like this.

Historically, a lot of them were too small to interest venture capital and too expensive for one developer to tackle quickly.

Now the economics are changing.

A business that might only support a few hundred paying customers can still be worth building if you can launch it quickly and operate it without a huge team.

You don’t need to invent the next Facebook.

You need to find fifty businesses with an annoying problem and build something they’ll happily pay to make disappear.

That’s a much less glamorous pitch than disrupting a trillion-dollar market, which is probably why nobody’s going to invite me to speak at a venture capital conference.

Fine by me.

You have an unfair advantage right now. It won’t last.

Here’s the bit I think people are underestimating.

The opportunity isn’t simply that AI makes building easier. It’s that AI makes building easier before everyone has adjusted to that reality.

Right now, you can still walk into an industry full of expensive, clunky software and build a credible alternative faster than many established companies can schedule a meeting to discuss whether they should form an AI working group.

Large companies have advantages. Distribution, existing customers, capital, sales teams and brand recognition. They also have legacy systems, approval processes, competing priorities and seventeen people who need to sign off on changing a button colour.

A small founder can move quickly because there’s nobody to ask for permission.

But this advantage has an expiry date.

The tools are going to improve. More people will learn to use them. Existing companies will adapt. Competitors will copy successful ideas faster. Customers will expect more features for less money.

And eventually, the same technology that makes it easier for you to build a product will make it easier for someone else to build a competing product.

That’s the catch.

If everyone can build software cheaply, the software itself becomes less of a moat.

Distribution matters more. Customer relationships matter more. Trust, domain knowledge, integrations, proprietary data and actually understanding the problem matter more.

The person who starts now has time to develop those things while the market is still figuring itself out.

The person who waits until the tools are perfect might discover that twenty other people have already built the thing they were thinking about.

And one of them has been talking to customers for two years.

Stop waiting for the perfect idea

I see developers do this constantly, and I’ve been guilty of it myself.

They spend six months thinking about an idea. They buy a domain. They redesign the logo fourteen times. They build a beautiful architecture diagram. They debate whether the backend should be Rust or Go.

Then they discover nobody wants the product.

Congratulations. You’ve built an exquisitely engineered solution to a problem that doesn’t exist.

The thing AI changes is how cheaply you can find that out.

Build the smallest useful version. Put it in front of real people. Ask for money. See what happens.

Not likes. Not upvotes. Not encouraging comments from other developers who will never use it.

Money.

Someone handing over their credit card details is a much stronger signal than a stranger on Reddit telling you the idea has potential.

And if nobody wants it, you’ve learned something without burning a year of your life and your entire savings account.

Move on. Try something else.

You can afford to be wrong more often now, which is an extraordinary advantage if you’re willing to actually ship.

Of course, that doesn’t mean you should throw fifty half-broken AI-generated applications onto the internet and call yourself a serial entrepreneur.

Customers still expect their data to be secure, their payments to work and the software to do what you promised.

AI can help you get to market faster. It doesn’t absolve you of responsibility for what you put into production.

The people mocking AI-assisted development are missing the point

There’s a particular type of developer who seems to believe that using AI to write code is somehow cheating.

They’ll proudly tell you they wrote every line themselves, as if customers are going to inspect the Git history before deciding whether the product solves their problem.

Nobody gives a shit.

Your customer doesn’t care whether you spent three days hand-writing a form component or had an agent generate the first version in ten minutes.

They care whether it works.

They care whether it saves them time, makes them money or removes something irritating from their day.

That’s it.

I love writing code. I still write plenty of it. I also like shipping products, and I’ve never met a paying customer who asked me to make their project take longer so I could experience the spiritual fulfilment of manually typing another API endpoint.

The craft still matters. Architecture matters. Testing matters. Security matters. Knowing when the AI has confidently produced absolute garbage matters quite a lot.

But confusing the amount of manual effort involved with the value delivered is a strange hill to die on.

Especially when the person you’re laughing at is already selling the product you haven’t started building.

This isn’t a guaranteed ticket to becoming rich

Before someone starts selling a course based on this article, let me ruin the pitch.

Most startups fail. AI hasn’t changed that.

You can build a product in a weekend and spend the next eighteen months trying to convince anyone to use it.

Distribution is hard. Sales are hard. Support is hard. Understanding what people will pay for is hard.

And having the ability to produce ten applications a month doesn’t mean you have ten businesses.

It might mean you have ten hosting bills.

There’s also a real possibility that software becomes so cheap to produce that customers become less willing to pay much for it. That’s already a risk for generic products with little differentiation.

The window I’m talking about isn’t a guarantee of easy money.

It’s a temporary opportunity to experiment at a cost and speed that would’ve seemed absurd a few years ago.

If you already know how to build software, you have a head start. If you understand a particular industry, you have another. If you can talk to customers, identify painful problems and actually sell something, you’re in an even better position.

Combine those skills with AI and you can punch well above your weight.

But you still have to do the work.

Sorry. The robot hasn’t figured out how to make customers care about your startup yet.

In a few years, everyone will wish they’d started now

I don’t know whether this window lasts twelve months, three years or longer. Anyone giving you an exact countdown is probably trying to sell you something.

But I don’t think the current imbalance between what a motivated individual can build and what the wider market expects will last indefinitely.

We’re at a weird point where the tools are already capable enough to be genuinely transformative, yet most people are barely scratching the surface.

That gap is an opportunity.

Not just for developers. For designers, accountants, mechanics, healthcare professionals, tradespeople and anyone who knows an industry well enough to recognise problems outsiders don’t even know exist.

You don’t necessarily need to become a world-class programmer. You do need enough judgement to understand what you’re building, test it properly and know when you’re out of your depth.

And you need to stop treating every idea like it has to become a billion-dollar company.

Build something useful. Find customers. Charge money. Improve it. Repeat.

If it fails, build something else.

I’ve spent a huge part of my career building software, and I genuinely can’t remember another time when the distance between having an idea and putting a working product in someone’s hands was this short.

That’s exciting. It’s also a little terrifying, because the same forces making this possible are going to make competition absolutely brutal.

Which is why I think waiting around for the technology to settle down is a mistake.

By the time everyone agrees AI is ready to build serious businesses, the easiest opportunities will already have people fighting over them.

The gold rush isn’t coming. It’s already happening.

And while everyone else is arguing about whether the shovel was made by a human, you could be out there digging.

Just make sure somebody actually wants the gold.