The Lever You Actually Have

It took me longer to make a cartoon dog's tail wag properly than to clone the software I pay $240,000 a year for.

That sentence is ridiculous. It's also the most honest thing I can tell you about building with AI in 2026, and this week I want to show you what it does to your bills.

Here's the one thing up front: you don't have to stop paying for software to stop overpaying for it. You just have to be able to stop. The rest of this issue is what "able to" looks like at one of my businesses.

Right now, while I write this, there are 16 windows open on the Mac in my office. Each one is an AI coding agent, a mix of Claude Code and Codex, and each of those runs its own little crew of helpers. We just counted the crews: 292 separate agents have touched this build so far, and the count went up while we were counting it. Together they're building our own version of the software that runs that business. The first working core got cloned in one night, on a free developer account.

Here's the part that sounds crazy: we may never use it.

We pay about $240,000 a year for the software it copies, and I want to be clear, I like that software. I built a lot of that business on it, and it made us better. This isn't a grudge, and it's why I'm not naming them.

But on our last call, the Chief Revenue Officer of the company that makes it, a publicly traded company worth billions, told me I might be "the canary in the coal mine." He said we'd be their largest customer yet to make this kind of jump, and that he's working on new pricing to keep us. We sit down Monday.

Either way, I win. A fair price on software we like, or we finish ours and run it. Both doors are good doors. When you can leave, every conversation changes, and you never even have to say it out loud.

And the same move has paid off multiple times this year. Here are three examples, and I can think of at least four more off the top of my head, but there's no sense listing them all.

The plan sets we have to file with the city used to come from an outside service at $80 to $200 apiece. We now make better ones in minutes, with the same team we already had, shaped around how our people actually work.

Our custom dashboards used to need an integrator, about $30,000 a year. The TVs in our office now run boards we built ourselves.

My favorite: a vendor came to pitch me a genuinely cool product. I told them before the demo that I'd never buy it, because I could build it. They presented anyway, and honestly, respect. Two days later we had our own version, and ours fits us better. That one kept $40,000 in the building.

I want to say the next part carefully, because it's the truth: nobody on the other side of those invoices did anything wrong, and some of their products made us better for years. The ground just moved under the whole arrangement, and it's still moving.

Now the part that keeps this honest: we don't always build.

We took a hard look at replacing the email tools that run two of our businesses, Klaviyo and Beehiiv. (This newsletter rides on Beehiiv.) We got far enough to prove we could, then put the wrench down. The juice wasn't worth the squeeze: their prices are fair for what they carry, and our hours are worth more than the difference. Part of having the lever is knowing when to leave it alone.

So how do you tell the bills apart? Here's the line I use: infrastructure versus interface.

Twilio is infrastructure. Pay them and you're renting carrier relationships, phone numbers, and delivery rates that took a decade to build. I can't clone that with 16 windows, and I wouldn't try. Shopify's lower tiers are the same story: payments, fraud protection, checkout, hosting, all for less than a dinner out. (Shopify also keeps wiring itself straight into the AI tools, which tells me they see what's coming. Time will tell.)

Interface is the other thing: the screens and workflows a vendor rents you on top of your own data. That's where the painful part of the bill has been hiding for twenty years, and that's the part AI just made cheap.

Want the strange proof? I'm one of the few people around who saves money by moving to Salesforce, because what I pay today costs that much more. Salesforce is infrastructure, and my screens go on top. Which makes me wonder who's out there right now building CRM infrastructure at Twilio prices. The day they show up, a whole category resets.

People stopped talking about the SaaSpocalypse a while back. It never ended. It just got quiet for a while.

Your steal this week: pull your last twelve months of software invoices and sort every line into those two buckets, infrastructure or interface. The infrastructure, pay it happily. The interface, pick the line that stings most and have AI build you the ugly version over a weekend. You aren't building a replacement. You're building a fact. Renewal calls go differently when you walk in holding a fact.

And if you can't tell which bucket a bill lives in, post it in the group. I'll tell you which side I'd put it on.

Danny

P.S. The dog's tail wags great now.