Willem Ave

So I'm going to use the days as the kind of where clause or whatever in the SQL statement and then pull back the data and kind of render it appropriately to the seller, to the customer.

I think it's an extension and just a realization of how to use.

For many years, I don't think we totally knew how to apply these things to real problems.

There was this very cool technology and there was this capability overhang that we just didn't know how to apply this perfectly.

Tool use and context window expansion, both of those you have to do together because you have to put all this data somewhere.

are really the two big unlocks.

And then the ability for these agents to work over a longer time horizon, right?

So 30 seconds, one minute, two minutes, five minutes.

It'd be totally reasonable for proactive insights for an agent to work for a couple of minutes or more, right?

Collecting data, testing it, making sure it's right before build a comprehensive report, kind of like deep research and others.

And I think those two innovations have really unlocked, I would think, far more

tangible business value than the kind of earlier, I would say, chat-based conversational interactions that were kind of the very beginning of this whole era.

My mental model now is more that these large language models are becoming more like the CPU of a lot of these different products that we're building.

And you need to give the CPU different things, right?

You need it to give it access to data.

You need to teach it how to use sandboxes and different tools so that it can effectively learn and get stuff done.

You need like some working memory, right?

You need all these like constructs.

And I think ultimately, even, you know,

Even if you ask someone to go do something, like they sometimes do the job wrong some percentage of time, right?