How Do You Balance AI Innovation with Enterprise Stability?

For us, building board agents that allow us a quick evolution without

destabilizing what the customer already built was one of the fundamental

constraints we had to put in place when we built the architecture.

The architecture is built on the Microsoft Agent Framework, so it's a product from

Microsoft Azure Stack that allows you to call multiple LLMs.

For us, it's important to be able to update our board agents and use the latest and

greatest frontier models underneath.

And Agent Framework allows us to do that with just configuration changes.

So whenever we migrate to a new LLM version, whenever something more powerful comes

out, we can adjust it. We can adjust the instruction without touching the planning

environment and the configuration of the planning environment our customers have in

place. Moreover, I would add that the boundary that we spoke about earlier, the

dataset API, which decouples the planning logic from the agent interfaces, well,

that interface allows us to roll out new agents without having an impact on the

underlying structure of data cubes and procedures and screens and all the other

traditional configurations that are present in a board implementation.

So we really wanted to introduce a layer to decouple the LLM configuration and the

interface of the LLM from anything else that you need to touch in a normal planning

implementation. This way, whenever we update the agents, we can make sure that you

don't have an impact on the rest of the planning environment.