Every time a customer messages Breezy to change a booking a series of small judgements happens in the background. Did they mean this Tuesday or next? Refund, or reschedule? Does this need a human? Behind every one of those judgements is a model doing the thinking and we've just changed ours.

‍

Breezy is moving from OpenAI's GPT to Anthropic's Claude. This post explains why we did it, and why we think it's good news for the way your bookings get handled.

Why does it matter?

You run a business. You don't care which model powers Breezy. You care that bookings get handled correctly, that awkward messages are interpreted kindly and that customers walk away feeling looked after. The model is the brain inside the product that ensures these goals are met. Changing it is a bit like changing who answers your phones. The policies don't change but the quality of every interaction does.

‍

When a customer writes 'push it to next Thursday, but maybe earlier', something has to read that sentence and decide what to do. That something is a large language model, or LLM the kind of AI that powers tools like ChatGPT and Claude.

‍

Reasoning on messy, real customer language

Customers write messages in ways we can't predict. As a human you unpack these odd sentences without thinking. Models vary in how well they do it. In our testing Claude handles this kind of ambiguity more reliably than the model we were using before. Fewer misreads means fewer back-and-forth clarifications and fewer frustrated customers. The real test of any AI isn't what it does when the path is clear. It's what it does when it isn't.

‍

A well-behaved model pauses and asks when it's uncertain. A less well-behaved one bluffs. This is the pattern often called 'hallucination', and MIT Technology Review has a good explainer on why it happens. When your booking calendar is at stake, a bluff is a mistake with a customer's name on it. From our side-by-side testing, Claude is more willing to stop and check than to push ahead on a guess. For a tool making real changes to your calendar on your behalf, that caution is exactly the instinct we want.

Why Anthropic, not just Claude

The model is half the story. The company behind it is the other half. Their work on AI safety is also unusually detailed for the industry. If you're curious, Anthropic's research hub is a good place to start. For a tool that makes changes to customers' bookings on your behalf, that transparency isn't a nice-to-have.

‍

When we pick a supplier, we look for focus. Anthropic builds models and the tools to deploy them safely. Their public roadmap is narrow and clearly communicated. That gives us confidence the technology we're building on won't lurch in a new direction and disrupt your service next quarter. We aren't knocking OpenAI. Their models powered Breezy for a long time and we're genuinely grateful for the work they've done. However, on the things that matter most to a booking product Claude is a better fit for what Breezy does today.

‍

Breezy is the carefully-wired AI agent that sits on top of your booking platform, with rules you can control and logs you can audit. Picking the right model is part of our work. Being honest about why we picked it is how we like to do business. If you want more on how we think about reliability in AI-handled bookings check out our blog posts.