Andy Halliday

And so these have more generalizable application, but the form and standards that are being created are really around coding at this point.

And so that's why the last piece is,

Goose is relevant because Goose is an agentic vibe coding capability that's completely local.

So it's running all on your local device.

It doesn't actually interact with the cloud in any way other than through MCP.

And it's a Gentic, meaning it's a coding system that can go out and access tools via MCP, but none of its logic or its brains, if you will, are actually consulting a cloud-based AI.

So those are the three components.

I just wanted to elaborate a little bit on what's forming there.

This is a very important announcement, and it's, I think, culminating the year of agents, as we wanted to call it.

And so at the very end of 2025, when agents have really become

you know, central to our use.

And I use an agent, GenSpark, every day.

It really is, you know, the new model for me for a really broadly capable and action-oriented assistant.

Well, now at the end of this year of agents, we have a new standard that's formed out there and a new organization to move forward with the development of agents that will be useful to you in many, many ways in the coming years.

So now for my next little tidbit.

Recently, DeepSeek, you know, the whale over in China, has released their newest model, V3.2.

And they make the claim that their V3.2 model, Speciale, that's the highest level of sort of reasoning and thinking model that they offer.

It has beaten Gemini 3.0 Pro on numerous benchmarks.

And so I wanted to talk about qualifications on that claim.

But the first thing to note is that they accomplished this pretty state-of-the-art, not state-of-the-art in terms of being the gold medal winner on any of these benchmarks, but they achieved this