
Meet Koa, Salesforce’s new AI brain
Salesforce just unveiled Koa, a CRM reasoning model for Agentforce that it built with NVIDIA Nemotron. Think of it like giving your enterprise software a better intern: one that can update opportunities, route cases, and schedule follow-ups without getting lost halfway through the task.
The pitch is simple: Koa is supposed to do more than answer questions. It’s designed to handle complex, multi-step workflows and pick the right tools along the way. Salesforce says it post-trained NVIDIA Nemotron 3 Super on a proprietary synthetic dataset built from nearly three decades of CRM know-how, then kept control of the model weights and inference inside its own trust boundary.
Why investors should care
This matters because Salesforce is trying to make Agentforce feel less like a shiny demo and more like infrastructure customers actually pay for. If Koa really cuts errors — Salesforce says it produced three times fewer on its CRM benchmark — that’s the kind of boring-but-valuable improvement that can turn into sticky usage and better monetization.
NVIDIA gets a nice cameo too. Even when the headline isn’t about chips, its AI plumbing keeps showing up under the hood like the stage crew at every blockbuster. More enterprise AI models built on Nvidia tech means more reasons investors keep treating NVDA like the toll booth on the AI highway.
The bigger picture
Salesforce stock was down a hair on the day, but the bigger story is strategic: the company is still trying to prove that AI can improve workflow automation without blowing up its control over customer data. If Koa becomes a real product customers trust, it could help Salesforce defend its turf against the “everyone has AI now” problem.
Big picture: this is less about a flashy chatbot and more about Salesforce trying to make AI useful enough that finance teams stop asking, “Cool, but where’s the ROI?”