And how do you, how do you get, I mean, how do you get those results with the model? Walk me through the, the, the mechanics here.

Yeah.

So, you know, the ability for us to really think about, you know, what, what goes into a harness, right? How does the model pull context? How does it use company data? How does it use its context window? How does it manage handoffs, right, uh, among tools? How does it apply guardrails? Um, you know, a harness is really, you know, all of those things and more.

Um, how it does QA, right? Um, how it spawns sub-agents, how it checks those sub-agents' work.

And when you, like us, really sit on top of, um, years of understanding how customers want to use, uh, agentic AI in, in the domain that we focus on, right, sales and marketing, um, it just gives you a tremendous leg up, um, in being able to really build, um, that, that model, build that harness.

The, the harness itself is actually model agnostic.

We have built it, um, to be model agnostic, but, um, and our platform is model agnostic, but of course, works best, um, with, with our model.

So you, you talked about OpenAI and Anthropic, and, uh, you had an interesting post yesterday on LinkedIn about, uh, watermarking, and, uh, I, I wondered if you could explain to us, what was the message you were trying to send with that post with respect to how some of these players are addressing watermarking, and then what that really says overall about their strategy around content and ownership and stuff like that?

So how do you balance it?

Well, we have policies in Rider, right? And this is, you know, the feature set that you build when you actually talk to customers versus talk at customers.

And those policies help an admin.

It doesn't have to be an IT admin.

Sales ops or sales enablement or marketing ops say, here are the ways in which I want my people using these connectors, right? And so you can say, you have to check an email before it sends out.

before it gets out.

You're not allowed to build automated email flows for one-to-one messaging, right? I'll give you an example from my workflow.

What comes to mind first?

Like why do we need any of these folks?

It is getting disrupted.

It's getting disrupted horizontally.

Let's start there, right? There is jurisdictional job overlap that is actually the, you know, where AI transformation goes to die.

If I am worried about taking over your job when I build my agent, I'm never going to build my agent as powerfully as I could.

So we actually, at the ground level, have to give people utter permission to rebuild workflows irrespective of how they've done it before or who has what job.

If you don't do that, you get a 54 human handoff process reduced down to 15 instead of zero or maybe one.

Can these large SaaS companies that have been pummeled by the market, can they reinvent themselves?

So bring us that nuance of when is it working when they're adopting writer either from the ground up or indeed within Salesforce offerings?

Yeah.

AI is not another software upgrade.

You can't outsource it to the CIO and expect results.

A lot of what we do is go to executives and CEOs and say, look, this is about you.

Sometimes you kind of want to shake them.

This has got to be a top-down, very heavy mandate to rewire processes end-to-end.

Are you seeing that coming from VCs just being more discerning, putting money for the winners, but taking away from those that aren't hitting scale?

Is this an issue that you guys have had to face at Ryder?

So there's no question that for the enterprise, right? What they want to be able to do is leverage what these super agents can do, right? And the most powerful thing that we can bring to a customer is the ability to bring down silos between teams, between systems, you know, between and across data.

I think there is absolutely a right to win for the systems of record around gentrifying access to and around their products, including to their own data.

I think, you know, rightfully so, folks don't want to, you know, create a system a subpar customer experience by having a ton of point solutions deliver thin functionality on top of what is essentially the data that they house.

At the same time, customers on the enterprise side who are spending tens of millions, hundreds of millions of dollars on systems of record want to make sure that you know, the agents that they build there and elsewhere have got access to all of the contexts that, you know, actually provide the value.

And so, you know, the way that we come to this is by really trying to be that trusted orchestration layer that delivers very, very high impact end-to-end workflows where the to win is obvious.

If I am combining call transcripts with CRM data, with deep web research, with proprietary information and databases, with business know-how and processes, right, that is just so far beyond what any one system of record can do.

Does it feel like that phase of AI to you right now?

When you onboard a customer, what does this look like?

Yeah.

So we are actually a verticalized platform.

And so when we onboard a CPG customer, it looks a little bit different than we onboard somebody in retail, onboard a technology customer.

We have libraries of hundreds of prebuilt agents, and they're quite powerful.

So, you know, a Qualcomm that does really sophisticated software ABM, that's account-based marketing, or a Salesforce that does incredibly powerful comms and PR, right? Their use cases are going to be different.

And what we're able to do is onboard them into libraries of pre-built agents that are ready for configuration for their systems, their tooling.

Can you talk about what these look like?

How do you get to that price point, Mae?

so it's really important for folks to understand what's required to get enterprise-grade agentic ai when we launched our ai hq a few weeks ago we had a hundred plus pre-built agents built for the enterprise and folks were asking me how are you getting this kind of efficiency and accuracy with really complex workflows agents at a uber at a franklin templeton things that were capable of taking multiple systems, data from third party and internal research, and able to generate really complex work.

We are able to take that full million token prompt and process that context window in 22 seconds, firing off multi-turn function calls in 300 milliseconds, all while getting this price efficiency.

It is a combination of breakthroughs, both algorithmic and data.

We train this model on about a million dollars of GPUs, so also incredibly efficient on the training, and we pass that on.

How

May, you talked about some of the very well-known companies that you currently serve, but to get these sorts of investors and the vanguards of this world to be coming on board, How are you making yourself known? I mean, are you having to offer this sort of price efficiency? Is that how you rise above the noise? Because it just feels like there are so many offerings, enterprise suited offerings out there right now.

It's absolutely so noisy.

Why is it very disappointing? Why is it so hard to bring this once-in-a-lifetime AI exuberance to bear on your productivity?

I think we are bringing knives to a gunfight when it comes to actually building with generative AI.

People are taking these really long kind of software development cycles, and IT is taking these projects to the business, and the business is saying, this is not really good enough to actually benefit my work.

So we take a very different approach.

Writer is... purpose-built for that kind of collaboration you need between the business and IT to actually deliver generative AI projects and applications and agents that work for the business.

So is it sort of legacy software is the issue here? So it's the legacy tooling of building these products.

And so the LLMs are very, very powerful, but you need to wrap it with data, with workflow, with know-how, and that's just not happening right now in most IT departments in the enterprise.

Can you articulate what that is?

You're going to have to rely if you're trying to get around export controls on setting them up, basically releasing them from other nations.

What we have been saying, Kelly, for two years is this big breakthrough, right? That you don't need billions to train state-of-the-art models.

We have proven that.

We built state-of-the-art models for less than a million GPUs, right? I don't think it is a reason to be bearish because we are on the cusp of an enormous agentic AI explosion that is going to require more inference than folks can even fathom.

Whether or not Trump tightens restrictions coming out of his meeting with Jensen today, the only way for the U.S. to stay ahead in generative AI is to continue to achieve revolutionary breakthroughs in training and model architecture.

The brute force method for training models is dead.

What's at stake here?

So in the enterprise, right, in the U.S., the trustworthiness of these models, the transparency of these models is of utmost importance.