← All notesSTRATEGY·23 Jun 2026·6 min

Claudini GPT: The True Bill

SaaS is dead, they say — just build it yourself. Sure. Here's the invoice the AI forgot to read you, from the $720-a-month bots to the divorce.

SaaS is dead. Why pay $20 a month when you can just build it yourself? I love this energy. I also ship software for a living, so let me read you the true bill.

Here's how it starts, and I know because it started with me. You're a writer. You're also, inconveniently for this story, an engineer. And you've used a few writing apps, and every one of them has the same quiet flaw: the notes, the world bibles, the references all live somewhere other than the page you're writing on. So you write a chapter, then you switch windows to check what colour your character's eyes were. Or you give up and keep it all in Apple Notes, and after the session you copy it back into the annotations panel like a raccoon hiding snacks. The discomfort is real. The diagnosis is correct. You understand the user perfectly, because the user is you.

So: build the app, right?

Right. And so you build the app. Claude Design even hands you these gorgeous blocks, and you sit there thinking, oh my god, it looks like something real — I'm about to become the Steve Jobs of editors and this is my iPhone 3G(S) (Product Red).

Well. Yes, but no.

Let's step away from the idea for a second and ask the unromantic question: what does that $20-a-month tool actually contain?

Claudini GPT: The True Bill
The true bill — everything the $20/month tool quietly assumes you already run.
  1. A front-end, where you click buttons and stuff. Got it.
  2. A back-end, that sends the emails, saves the precious manuscript you're writing, and makes sure it does not accidentally land in the account of your weird neighbour — the one writing a 90,000-word memoir about three months of growing cherry tomato seedlings and how the other neighbour's cat ate them. The seedlings, that is. Not the tomatoes.
  3. A database, where the seedlings and the cats are stored.
  4. Billing. Recurring billing. Secure handling of the billing.
  5. GDPR, DPF, and the rest of the paired-letter alphabet.
  6. Caching, a CDN, and Redis because someone in a forum said you need Redis.
  7. A domain, MX records, DMARC, SPF. (The Sender Policy kind. No, we're not going to the beach.)
  8. An SMTP relay, or a service that is one.
  9. The nineteen pages that aren't the editor: billing, profile, your books, account, security, report-bad-content, and 2FA management — which is its own page, which means it's also number ten.
  10. So: 2FA, plus OAuth, so the user can log in with their [email protected] account. Don't forget email validation.
  11. Antispam. Bot prevention — to stop the other $20/month invention from registering you 972,334 fake accounts (that's roughly an extra $720 a month in DB & storage, by the way).
  12. And then the part nobody puts on the list: setting up the server, securing the server, and the deploys. The unglamorous spine of the whole thing — the part that's actually hard, and the part everyone assumes is free.

I could keep going. You get the idea.

Now, back to the app. You're holding a pile of deeply complex machinery that you have no idea how to verify — not whether it works well, but whether it works at all. Call it $400 a month in servers and maintenance, plus the $720 once the bots find you.

And it's been three months. No weekends. No evenings. And now — no wife.

So you're down $400 a month, roughly $300,000 in separation and alimony (don't forget the alimony), and your dog needs a $2,000 procedure because the divorce stress finally reached him too.

All to save $20 on a subscription that an AI told you would be extra easy to build yourself.

A multiplier, not a replacement

Here's the thing. AI is not a replacement for expertise. It's an expertise multiplier. If you already know how to build software, you build it faster. You write the tests you'd otherwise have skipped — you have them at all, because you're not selling the house to afford the time. You generate the report. You process the data. You finally find the spreadsheet bug that's been quietly robbing you since four accountants ago.

It's still you. You didn't know how to stand up a secure, high-load-ready server, and you still don't. Fine — you heard the word “DevOps” somewhere and asked Claude to handle it. I'll bet you $20 it breaks on the first deploy. And the second. And the third. And the seventh.

It breaks on the first, second, third, and seventh deploy for real DevOps engineers too, and that's normal. The difference is that they know it broke. You just see that the site isn't there, while Claude cheerfully asks whether you'd like the Ansible cookbook updated or to switch to Kubernetes.

So, as with a lot of cool things in life, it's not what you pictured. And that's fine.

That's why it's probably fine to pay $20 a month for a tool built to do one specific thing, that's been around a few years, that outgrew its software adolescence and now — really, more rather than less — just works. It doesn't do everything. It takes a few real burdens off your shoulders, and you're a decent expert at your actual profession, and AI makes you better at it.

So in the end it's easier to pay the $20, keep doing your thing, go to the gym, drink the wine, hug your wife.

The most expensive line was never on the invoice. Buying the tool means renting someone else's command of all that context. Building it yourself means becoming that context. Permanently.

That's the bill nobody reads back to you.

Unless, of course, you really wanted the divorce.

Still with me?

Weighing a rebuild, a rescue, or a roadmap?