Code became free. Building a real business didn't.

This week I spoke to six founders who told me they’d “launched” a SaaS with Claude Code or Cursor.

Five of them have a working prototype. None of them have a real product in production with paying customers yet.

And it’s not a technology problem.

The 2026 vibe coding pattern

For eighteen months now, I’ve been watching the same sequence play out across very different entrepreneurs — first-time founders, ex-corporate tech employees, operators who wanted to prove they could build a product themselves.

You start on a Saturday with Claude Code or Cursor. Within hours, you have a UI. Within days, you have a working user flow. Within two weeks, you have a prototype that would make whatever you built in 2018 over six months look like a hacked-together demo.

The high is real. You finally see the thing you’ve had in your head for years without being able to code it. You tell yourself: this time, I’m going to do it.

Then you try to put it in front of real users. And things get complicated.

The gap nobody pitches in a demo video

Between a prototype that wows and a product you can actually sell to B2B companies, there’s a list of topics no vibe coding video shows you.

Security. Not the “I added HTTPS” kind — the real kind, the kind your first enterprise client will audit. Permissions, logs, authentication, secrets management, the injection attacks your generated code didn’t anticipate.

Deliverability. If your product sends email, get ready to spend six months understanding SPF, DKIM, DMARC, IP warming, blacklists, and why your emails go to spam for 40% of your users.

Uncontrolled behaviors. This is the topic that worries me most about companies I see building with AI agents in production. When the LLM decides something you neither tested nor anticipated, the responsibility is yours — regulatory, contractual, sometimes personal.

Support. Legal. GDPR. Billing. The edge case bugs that only show up at 200 simultaneous users. The integrations your client asks for on the second call.

None of these topics is blocking for a prototype. All of them are blocking for a real product.

Why most founders give up at this point

It’s not the tech that stops them. Claude and Cursor can help on 80% of these topics.

What stops them is the sum. The cumulative weight of everything you have to solve, on top of the product itself, just to be taken seriously by a real company.

And here, I see two scenarios.

The first: the founder disperses. They start coding their own auth system, then a custom billing engine, then a Salesforce integration, then an analytics dashboard. Six months later, they’ve rewritten half of Stripe and Auth0 — and their product still hasn’t found its market.

The second: the founder quietly gives up. They keep the project as a side project, go back to their main job, and end up buying an existing tool to solve the problem they claimed to be solving. Which is, let’s be honest, often the rational decision.

But the pattern, in both cases, isn’t a technical problem. It’s a focus problem.

What I’d have underestimated about Sortlist

When I started Sortlist, if anyone had shown me a quarter of half of what it would take to reach €10M ARR, I’m not sure I would have started.

Not the code. The code was never the problem — not in 2014, even less in 2026.

What I would have underestimated is everything else. The nine countries to open one by one with different local specifics. The three economic model pivots we’d have to make. The hires I would get wrong. Compliance in every jurisdiction. The client contracts that take six weeks to sign because procurement asks for an ISO audit. The boards where you have to defend decisions you find shaky yourself. The eleven years of consistency.

No tool reduced that part. Not in 2014. Not today.

And that’s exactly the trap of vibe coding right now: we’ve massively reduced the cost of the first step — building the product — and that creates the illusion that we’ve reduced the cost of all the steps. We haven’t.

The real cost of a business hasn’t changed

Code has become almost free. Building a real company costs exactly what it always cost.

What costs now isn’t writing the first version. It’s:

Finding the first ten paying clients who actually stick
Understanding precisely the problem you’re solving, for whom, and why
Holding focus on a single thing for 18 to 36 months without dispersing
Building the operations, support, and compliance processes that make a company stand
Hiring the right people at the right time, and not three months too early
Saying no to 90% of the good ideas so you can execute 10% properly

None of these are addressed by Claude Code, Cursor, or an AI agent. They’re focus, discipline, and patience problems. The kind of problem founders have always struggled with — and that the ease of prototyping makes, paradoxically, even harder.

Because when every new idea can be prototyped in a weekend, the temptation to prototype them all becomes irresistible. And you end up with a graveyard of prototypes instead of a product.

The principles that still hold in 2026

Three old, sometimes mocked principles have become more important than ever.

Focus is king. One client target. One problem. One use case you nail completely before looking sideways. Every founder knows this. Almost none actually does it, because saying no to 95% of opportunities requires a discipline you can’t learn from a book.

Ship more, ship faster. Not to produce more features. To learn faster. Every release you put in front of real users gives you more information than three weeks of planning. Vibe coding makes that loop shorter than ever — provided you don’t confuse “I have a prototype” with “I shipped.”

KISS — Keep It Simple, Stupid. In 2026, this has almost become an act of resistance. When AI lets you code a feature in an hour, the trap is to ship ten of them before validating that the first one was right. The product that wins in B2B is still the one that does one thing better than everyone else — not the one that does fifteen things mediocrely.

What I’d tell a founder starting in 2026

Three pieces of advice, in this order:

One — Pick a problem you live personally, or that a real client already pays you to solve by hand. Not an idea you had in the shower. Not a problem you saw in a LinkedIn thread. A problem whose exact cost you can describe for a real client, in euros per month.

Two — Build the worst possible version that solves that problem. Truly the worst. An ugly UI, missing features, manual integrations. But one that actually solves the problem for one specific user. If you’re embarrassed to show it, you’re probably just minimal enough.

Three — Find ten paying clients before writing the second version. Not ten free users. Ten clients who pull out a credit card. If you can’t reach ten after three months of serious effort, the problem isn’t the product — it’s the problem you chose.

Vibe coding doesn’t exempt you from any of these three steps. It just lets you do them with less money than before. That’s a massive advantage if you have the discipline. It’s a trap if you don’t.

The short conclusion

The bottleneck of a tech company has moved.

In 2014, it was often the code. In 2026, it’s almost never the code anymore. It’s the focus, the discipline, and the ability to say no to every feature your tool lets you build in two hours.

AI democratized building. It didn’t democratize business. And most founders launching today will learn that lesson the hard way — not because they can’t code, but because they haven’t yet realized that coding was never the point.

If you’re starting today: prototype less, ship more, and pick the one problem you’re willing to defend for the next five years.

The rest is noise.

Thibaut Vanderhofstadt is co-founder of Sortlist — Europe’s leading B2B matchmaking platform, active in over 140 countries. He now works with SaaS founders, marketplace operators, and ambitious non-tech companies through MetSaaS. Book a diagnostic →