Willem Ave

And what does that do to my tips?

on Thursdays or whatever, I don't want to discount how well they know their businesses.

So there's very specific things they want to find out to help them run a better business.

And I think the interesting next step is how we can be more proactive and proactivity for suggesting things that they may not be thinking about.

And this is also when you combine these large language models with tool use.

Right.

You can actually get very grounded, correct things to help them understand what's happening.

And that's like a really key investment area for both us and most of the industry.

This is part of the art of building, I would say, products in today's age, right?

So there's a few things like kind of when first ChatGPT first came out and a lot of these large language models came out, there was a lot of quote hallucination, right?

You would ask these things, they'd just be like blatantly not true.

The real innovation, in my opinion, is one, yeah, they've gotten better at pre-training and better kind of data inputs.

But the second really is that we've taught, you know,

basically these large language models to do better agent loops.

This is one of the reasons why Gemini and ChatGPT are much better now is that they integrate search data.

They can enrich their context with actual data from tools.

If you think about the specific example on Rainy Day Tips,

What we do is we have an MCP ecosystem that we connect our AI tools to, and there's a weather MCP effectively, and there's a data warehouse that these language models actually write real SQL against.

It's like as deterministic as possible.

And then basically you can kind of reason through it that, you know, get the rainy days, those are actual days.