Multi-Agent Systems: Orchestrating Intelligence with Domo and Snowflake Cortex AI | Domo BUILD 2026

Excited to be here. Thank you to the Domo team for the invitation. Good morning, good evening, good afternoon to all our viewers. And my name is Mayur Mahadeshwar. I'm the Partner Sales Engineer for Domo at Snowflake. Work very closely with the Domo team on a lot of our mutual customers, deployments, engagements. So, you know, very excited to be here today and what we are talking about is multi agent systems.

Now there has been a proliferation of agents, but a lot of business users like and love the Domo interface, but they do have data agents which they have already built in Snowflake. So today I'm going to talk about how you can use Domo and it's excellent AI abilities for orchestrating agents. You ask questions of your Domo agents, but then the Domo agents actually use Snowflake agents underneath the covers.

So just a high level overview of the Snowflake architecture. At the lowest level we have the context layer. On top of that, you build your Snowflake agents and on top of that you have your experiences via Snowflake Co work or Snowflake Coco.

So the real challenge as I mentioned earlier was if you have worked with LLMS, you realise that LLMS are not deterministic. They are most of the times they are probabilistic, which means that you want to make sure that the answer that you get from your data platform like Snowflake is the same answers that you get when you query from a tool like Domo. So in order to do that, you want to make sure that you are as often as possible reusing your agents and some of the task which you have, you know, offloaded to your data agents hence return the same responses.

So what we have over here is a multi agent pattern that I'm seeing more OFT used across customers. In this case, a user logs into their Domo environment and uses Domo as the main orchestration or the interface for doing various tasks. And the Domo agent essentially is the brain of the system. The system uses LLMS. You can either use Snowflake as the inference point or you can use, you know, any other model provider of your choice.

But the agent or rather the Domo agent then decides, hey, this is a task that is probably best done by the Snowflake agent that I already have. So let let me use and let me offload this task to a Snowflake Cortex agent. The Snowflake Cortex agents has a set of tools which are specific to Snowflake for that task. It could be your Cortex analyst, Cortex search your semantic views. So that way you don't have to worry about having, you know, like different responses across different sets of data.

So now the snowflake agent basically runs that task and then returns the result back to your Domo agent, who then based on that result can maybe trigger a workflow, take any action, or maybe even for that matter, call another agent from another data platform.

So that's how a pattern for multiple agents works and it can get, you know, pretty complicated and there could be like multiple levels to this. So the snowflake agents can then use a swarm of agents to accomplish the task. The possibilities are just endless over here. Now, enough of the slides. Let me actually show you all this in action.

So I have this Snowflake agent over here which is essentially working off a set of clinical studies. Now these are all clinical studies which exist in Snowflake and I have built an agent on top of it. Now, if I ask this agent any question, I do get a response, right? How many studies are in phase two? Let's see.

So it returns the number of studies which are in phase two. This is good. But what if I want the same response when I query it from Domo from a from the same user persona. So I have the role as a Domo role. So I have this agent in clinical agent in Snowflake. Now, whenever I use this agent, I want to make sure that you know how many studies are in Phase 2.

So I want this agent to actually offload the task to the Snowflake agent. As you can see over here, it's actually delegating the operation to the Snowflake agent that we have defined in Snowflake. This is our Snowflake agent that's actually doing the work now, while it's actually returning the result. You can also check the query that Snowflake ran for this for for getting the result and the query is available over here.

Now let's go back to Domo and see if we have the same results. So as you can see over here, we have the same 13 studies which are currently in phase two. And the answer is consistent whether I query, yeah, you know it from Snowflake or whether I execute it from Domo. And that's the sort of consistency that I am looking for. I want to make sure my business users and my technical users get the same response.

And just to show you that, you know, we are actually, there's also observability on your agents. So just to show you that you're not actually cheating over here, but these are the two request ID which I just ran. And as you can see, one is originating from an external application, the other is coming from Snowflake Intelligence or Co work. We have rebranded Snowflake Intelligence to be Co work now. And this is the exact same question and the same agent responses.

So this way I can assure that the multi agent system that I have built is intelligent enough to offload specific task to the data agents underneath it and then surface the results on the Domo side. That was my talk. Thank you so much for all and hope you have a good build.

‍