We cut our AI bill 74% in 3 months

We cut our AI bill by $100k a month, but the real win was never the money.

Amos Bar Joseph September 11, 2026

I’m Amos Bar Joseph**, co-founder of Swan, the first Autonomous Business OS. At Swan, we’re building what we call the Autonomous Business: a company that scales to $10M ARR per employee with no bloat, no assembly lines, no Cog Culture. Just humans in their zone of genius, amplified by AI agents.*** I write The Autonomous Age to share contrarian insights from that journey, on GTM, leadership, and the future of work. If you want to understand how GTM evolves beyond playbooks and assembly lines, this is where the story unfolds. Connect with me on Linkedin or X. If that's not the game you're playing, reply to unsubscribe.

Our anthropic monthly bill went from $135k to $35k in 3 months. ARR almost doubled in the same period. Our goal? get it to ZERO by end of year.

It all started three months ago. I was staring at that $135k invoice and had a realization that had nothing to do with cost.

"Anthropic is not building for us anymore."

No drama in that sentence.

Like everybody else, they have an ideal customer profile.. and we're just not in it.

It all became crystal clear when I started looking at Anthropic as my landlord, and my app as the tenant.

You see, at Swan AI** we were paying $135k a month in rent for a good building.

The landlord kept renovating, and every renovation came with a higher bill.

Then one day we looked around and realize none of the renovations are really for us.

The cheap units stopped improving.

The penthouse kept getting better, and kept getting pricier.

That's the whole realization: anthropic's models are built for Claude, and Claude is built for huge enterprises doing heavy software development, finance, insurance.

not our space.

And if you're building an AI product on top of their models, probably not yours..

So we started moving out. early, before it was obvious. we called it project Exodus.

Exodus started when the open source models took their first big shot at the frontier. Minimax dropped. GLM right behind it. We started running the evals expecting a gap..

But there wasn't one. Not for the work that fills our customers' days.. scoring leads, reaching out to prospects, updating the CRM, summarizing calls.

We got the same output, at a fraction of the cost.

So we rebuilt the stack around where the work actually goes.

Open source runs the day to day. Anthropic only gets the handful of tasks that are genuinely complex.

Performance didn't move. The bill did.

But then something even bigger happened.. we started owning our destiny.

We felt like we were no longer building in the shadow of uncertainty, waiting for the next model release to see if it kills our business or not..

The moment we started owning our tokens was the moment we took control over our business.

Now I'm not telling you to own your model.

It's complex, and almost nobody should do it today.

I'm telling you to have a STRATEGY where owning doesn't look impossible.

What if training your own model is about to become absurdly simple? So simple it's close to a click of a button?

Would that kill your business, or make you unbeatable?

*[getswan.com

](https://getswan.com/?utm_source=theautonomousage&utm_medium=referral&utm_campaign=we-cut-our-ai-bill-74-in-3-months)

A vendor's pricing and priorities can shift overnight, and the only real protection is having a strategy ready before that day arrives, one where owning an alternative doesn't sound impossible.

The dependency audit*

• List every AI vendor your roadmap depends on, and mark which ones you could not operate without tomorrow.

• For each one, write down what breaks if their pricing or priorities changed overnight.

• Move whatever is routine and high volume to the cheapest model that still gets it right.

• Keep the frontier model on call only for the handful of tasks that actually need it.

• Write one sentence: what's your move the day owning your own model gets easy.

Which part of your stack would you most regret depending on if your vendor's priorities shifted tomorrow?

The strategy matters more than the savings, because the day owning an alternative gets easy, you want to already know your move.

-Amos

Read more.