$113K: The invoice that went viral and the story nobody told correctly.

Reddit called it front page worthy with 26,000 votes.

Business Insider published a full piece.

Half a million saw it here.

The comment sections had opinions.

All of them were doing the same autopsy. On the wrong body.

The question everyone asked was identical: how does a 4-person team justify a $113K AI bill? Business Insider framed it as a margin question. Reddit framed it as either recklessness or proof of concept. The journalists wanted our revenue numbers. The skeptics wanted us to feel embarrassed.

Everyone started at the number. Nobody looked at what it reveals about the industry.

$217K in AI costs across four months. Four people. 800+ companies running workflows through the platform last month.

That number wasn't a surprise.

When 800 companies run GTM workflows through your platform, every execution burns real compute. The Observer monitoring LinkedIn signals. The Hunter identifying website visitors. The Gatekeeper qualifying inbound leads. The Listener analyzing sales calls. Every single interaction, tokens. The bill isn't a warning, it's a readout of the machine working.

But here's where it gets interesting.

This isn't software. Software gets cheaper as you scale. More customers, lower cost per customer, expanding margins. The SaaS dream. AI doesn't work that way. AI costs move with your customers. Not against them. The bill grows because the business grows. Those two things are the same thing.

So when that bill arrives, every AI company faces the same pressure. And most make the same call.

Mark up the tokens. Bundle them into pricing. Buy wholesale from Anthropic or OpenAI. Sell retail to the customer. Take margin on every token consumed.

Clean. Logical. Obvious.

Also a trap.

Because you're not selling your product. You're reselling someone else's. And the moment you do that, you've handed your competitors the simplest playbook in the world. A competitor shows up willing to take a thinner margin on the same underlying technology. Not your technology. Anthropic's. Then another. Then another. Suddenly the conversation isn't about what you built. It's about who blinks first on token margin.

You're not competing on your product anymore. You're competing on how little you'll eat on infrastructure you didn't build, don't own, and can't differentiate.

There is always someone willing to go lower.

That's not a business model. That's a race to the bottom.

So where does that leave us?

We made a different call. Compute is a utility. Treat it like one. Pass it through at cost, transparently. Not a profit center. Infrastructure. The electricity that powers the building.

What you charge for is the building.

At Swan, that's the orchestration layer. The system that takes a GTM motion described in plain language and turns it into automated workflows running without a team behind them. The agents that know which signal to act on, when to act, and how to act. That's what we built. No one undercuts us on it by squeezing Anthropic's margin, because it isn't a commodity. It's ours.

And here's the part that really clicks. You could let customers bring their own models entirely. Found better pricing elsewhere? Plug it in. It doesn't change the equation. Because what they're paying for was never the intelligence sitting underneath. It was always the harness around it.

Here's what's funny about all of this.

It's not a new idea. It's not even a brave idea. It's SaaS economics. Charge for the software you built. Let the compute be infrastructure. We've run this playbook before, with every layer of infrastructure that felt transformative until it became a utility. We figured this out before. Then AI came along, made everything feel unprecedented, and somewhere along the way everyone forgot what they already knew.

The ones who remember it first are building something real. The ones who don't are building a margin war they will lose.

Run the harness audit before your next pricing conversation

• Open your pricing page. Draw a line between what you're charging for that you built, and what you're reselling. If that line is hard to draw, that's your answer.

• Run the model substitution test. Ask: what happens to your business if Anthropic cuts token prices by 50% tomorrow? If your margins improve automatically, you're charging for the harness. That's where you want to be. If you have to reprice everything, you're in the token game and the clock is already running.

• Bring-your-own-model question. Could a customer plug in their own model and still find your platform worth paying for? If that question makes you nervous, the harness isn't built yet. If the answer is obviously yes, you already know which business you're building.

The $113K isn't the number worth watching. The harness around it is.

–Amos

Amos Bar-Joseph

Founder & CEO of Swan AI. Building the first autonomous business scaling to $10M ARR per employee with AI agents. Writing about the future of GTM and leadership.