Where is all the money going?

The honest story of AI build credits — why they exist, what actually powers them, and why the cost curve flips in your favour once your site is built.

It's a fair question. You've used Wix. You've used Squarespace. You dragged things around, hit publish, and the whole experience cost you nothing upfront. So when Ullbek asks you to spend credits to build, the natural reaction is: wait, what? Where is the money actually going?

Here's the completely honest answer. No marketing spin.

First: understand what drag-and-drop actually is

Drag-and-drop builders are free to use because nothing new is being made for you. That's not a criticism — it's just the architecture. Every template you pick already contains every line of HTML, CSS, and JavaScript your site will ever use. You're not building anything. You're filling in blanks.

Drag-and-drop builders

Assembly, not creation

Template ships with all the code pre-written
You choose colours, images, and text from presets
Software records your choices — nothing is generated
Free because the hard part was done years ago, by someone else
Ullbek (AI build)

Creation, from scratch

Every message triggers original code being written for your brief
A frontier AI model reasons about your layout, typography, and intent
Real HTML, CSS, and JS — produced on demand, unique to you
Credits cover the cost of that inference — we pass it through at near cost

Think of it this way: drag-and-drop is like choosing a car from a dealership lot. Ullbek is like commissioning a car to be built to your specification. The dealership can offer the lot car cheaply because it already exists. The commission costs something because it's being made.

What actually happens when you spend a credit

A credit is not a subscription fee, a storage fee, or a margin. It's the cost of running a powerful AI model through your build request — passed to you at near the wholesale price we pay. Here's where it goes:

Where your credit goes — the full chain
💳
Your credit
You spend one credit per build action
🧠
LLM inference
The bulk — frontier model reads your brief and writes code
⚙️
Agent layer
Browser verification, error fixing, design context
🌐
Your site
Original code, live and owned by you

We are not making meaningful margin on credits. The AI model we use is expensive. We chose to absorb that pricing structure rather than switch to a cheaper model — and there's a specific reason for that, which brings us to the part most people don't think about.

Why we can't just use a cheaper model

This comes up a lot. There are open-source models. There are cheaper API options. Why not use those and keep costs down?

We tried. We tested a lot of them. And what we found was consistent: cheaper models produce code that works, but rarely produces a website that looks beautiful.

"Building a good website is not a mathematical problem. It's an aesthetic one. And only certain models have developed the taste to solve it."

LLMs are extraordinarily good at maths, logic, and syntax. But design is different. The difference between a website that looks professional and one that looks amateur is almost never in the code structure — it's in the margins, the spacing, the font weight choices, the colour contrast, the visual hierarchy. It's in a hundred small decisions that require taste, not calculation.

💸
Cheaper models

What they get right

Claude Sonnet

What they add

  • Valid HTML, CSS, and JS
  • Correct layout structure
  • Reasonable component logic
  • Inconsistent spacing rhythm
  • Poor typographic hierarchy
  • Flat, template-feeling results
  • Everything above, plus:
  • Nuanced spacing and proportion
  • Intentional typographic scale
  • Aesthetic cohesion across pages
  • Design taste in ambiguous situations
  • Results that feel crafted, not generated

The devil is in the details — and on the web, those details are everything. We experimented with cheaper models for months. The honest conclusion: they can build functional websites, but they cannot reliably build beautiful ones. And at Ullbek, we made a decision that we would never publish something we'd be embarrassed to show a designer.

That's the premium you're paying for. Not compute. Taste.

The cost curve actually flips over time

Here's the thing people don't realise on first encounter with Ullbek: the credit model feels expensive at the start, but it's structured in a way that rewards you over time.

Think about how websites go stale. The biggest problem with the existing web — tens of thousands of small business websites — is not that they were built badly. It's that they were built once and never updated again. Changing a price, adding a page, updating the menu — all of that requires either paying a developer, or logging back into a drag-and-drop editor and wrestling with something you barely remember how to use.

Building the site

The big spend

The AI writes everything from scratch — every page, every section
This is where most of your credits go, and where most of the value comes from
Think of it as your one-time custom build cost
After it's built

Tiny, cheap updates

"Update the opening hours." — a few cents worth of tokens
"Add a page for our new menu." — done in seconds, costs almost nothing
Because it's this easy, you'll actually keep your site up to date

With Ullbek, once your site is built, maintenance becomes extraordinarily cheap. Because all you're doing is prompting a change — a single sentence, a small instruction. The tokens spent on "add a page for our new menu" or "update the catering price to £45" are a tiny fraction of what the full build cost.

What this looks like in practice
  • Initial build: You describe the whole site. The AI writes everything from scratch. This is the heaviest credit spend — and it's also the most value you'll ever get from a single session.
  • Ongoing changes: "Update the opening hours." "Add a gallery page." "Change the hero headline." Each of these costs a tiny amount — just the tokens for that one instruction.
  • The result: You actually keep your website updated. Not because it's free — but because it's easy enough that you actually do it.

The biggest problem in the existing web ecosystem isn't bad websites. It's stale ones. Sites that were built in 2019, haven't been touched since, and are sitting there with wrong prices, old phone numbers, and pages that no longer make sense. Ullbek is the first builder where maintaining a site is as easy as sending a message.

The short version

If you want the condensed version, here it is:

The honest summary
  • Credits exist because real AI inference costs money. Every build is a live call to a frontier model. That call is not free, and we don't pretend it is.
  • We use an expensive model on purpose. We tested cheaper options extensively. Only the best models produce websites that are genuinely beautiful — and that's the whole product.
  • You're not paying for software. You're paying for a build. Think of it as commissioning work, not subscribing to a tool.
  • The big spend is upfront. The initial site build is where most of your credits go. Every change after that is a small, cheap prompt.
  • The payoff is a site that stays alive. Because updating it costs almost nothing, you'll actually do it — which is the thing that makes a website valuable.

We'll cover the exact numbers — how much a realistic full build costs, how much a monthly update costs, how it compares to a developer or a designer — in a future post. For now, we just wanted to be honest about what's happening under the hood.

Because you deserve to know what you're actually paying for.

Start with $1 in free credits

See what a full build actually costs. No card required — your first dollar is on us.