Hey!

Chris here. Welcome to Blueprint—the newsletter to help you build a winning engineering team.

In the last 2 weeks, I've had the same conversation at least 10 times.

Different companies of different sizes, working in different industries, but all asking the same question: "We're spending a lot on Claude and OpenAI. Is this just what AI costs now?"

The answer is no. And the reason why says a lot about what companies are getting wrong with their AI strategy right now.

Let me explain. 👇️ 

📒 DEEP DIVE

Claude is Not an AI Strategy

Why single-vendor AI stacks cost more, do less, and leave the most important problems untouched.

One Ecosystem, One Set of Rules

Let me acknowledge the obvious upfront: Claude is a genuinely good tool.

Anthropic has built one of the more capable, business-friendly AI ecosystems available, and if you're using it, you haven't done anything wrong.

The problem shows up when companies treat it as a complete AI strategy rather than one piece of a much larger picture.

The core limitation is that Claude is built on a single LLM and structured around one company's rules about what you can and can't do.

For experimentation and casual use, that's fine. But when you try to build real workflows—automating meaningful parts of your business rather than just asking questions—those rules become an obstacle.

There are things Claude will simply refuse. You work around it, hit another wall, work around that. Anyone who's tried to build something substantive inside that ecosystem has run into this exact pattern.

That's a frustrating constraint on its own. But the economic problem underneath it is worse.

Don't Ask a Barber If You Need a Haircut

Anthropic makes money selling tokens, and the whole system is built around that. So when you're relying on Claude to guide how you should be using AI, you're asking the person selling the product to advise you on how much of it you need.

This is exactly why CEOs keep calling me. They've looked at their bills and done the math. Some of them are seriously asking whether they'd be better off going back to paying humans for certain work, because the numbers are starting to look close.

When you get to that point, something has gone wrong.

Here's what they're missing: with better systems—systems that aren't locked inside a single vendor's ecosystem—you can bring those costs down to roughly ¼ of what they currently are, using the same underlying models.

The leverage is in the harness around those models, and who controls it.

Claude's ecosystem makes that flexibility nearly impossible. Their API tokens are expensive enough that the cost math stops working at scale, and routing to other models or managing your own usage patterns requires significant effort to work around.

Most companies don't have the bandwidth to fight that battle, so they just pay the bill.

The Use Cases That Fall Outside the Box

Beyond cost, there's a second problem that's less obvious: there are entire categories of AI use that simply aren't in the purview of what any single model company would do for you—and probably never will be.

Think about the software that actually runs your business:

• Your ERP

• Your financial systems

• The operational workflows your team has built over the years.

These are the places where automation would change how your business operates day to day. But a generalist chat interface wasn't designed for any of this, and the more locked-in your AI stack is to a single vendor, the harder it becomes to build anything that reaches inside those systems in a meaningful way.

To automate the parts of your business that actually move the needle, you need a system flexible enough to look under the hood. One that can be configured with access controls, pointed at your actual data, and made to understand your specific business context—not just respond to prompts.

That kind of depth requires a flexibility that a closed, single-vendor ecosystem just can't give you.

How To Do This Right

What works is walking a business end-to-end—from intake through operations through invoicing—identifying where the real friction is, and building something specific to that friction. Find the prickly parts. Configure a system to handle them.

The model I've come to believe in is services paired with software.

Setup has to be fast and affordable, but you need someone who actually learns the business, maps the context, and applies that knowledge to configuration.

The feedback loop between the people building the system and the people using it is where real improvement compounds over time.

BEFORE YOU GO…

The CEOs calling me about their AI spend aren't careless. They used the most accessible tools available, which made complete sense when they started.

Accessible tools, though, rarely optimize for how a specific business actually runs.

Large companies have a hard time with this model—too much red tape, not enough ability to move fast. The real opportunity here is for smaller companies that can make a decision on Monday and have something running by Friday.

And here's the part that doesn't get said enough: Anthropic and OpenAI aren't coming for that market. The dollar opportunity isn't big enough to be attractive to them at their scale.

Which means the businesses best positioned to build a real advantage here are operating in a space the biggest names in AI have looked at and passed on.

Talk soon,

Chris.