Salesforce is expanding Agentforce across AWS and Google Cloud while developing a specialised CRM reasoning model alongside NVIDIA. This broad technical push links its AI agents more closely with underlying enterprise data, daily workplace applications, and core cloud infrastructure.
Announced at Dreamforce 2026, the coordinated updates span multiple layers across Salesforce’s AI stack. The newly introduced Koa provides the company with its first reasoning model developed specifically for CRM work, enabling software to carry out complex enterprise processes. Concurrently, fresh AWS integrations bring Salesforce data into Amazon Quick and embed AWS agents into Slack, while the expanded Google Cloud partnership delivers robust infrastructure, Gemini Enterprise connectivity, and new commerce capabilities.
Koa is trained for multi-step CRM work
Built on NVIDIA Nemotron 3 Super, Koa serves as Salesforce’s first dedicated CRM reasoning model for Agentforce. The system was post-trained using synthetic data modelled on enterprise knowledge accumulated across nearly three decades of real-world CRM deployments. These simulated environments cover more than 14 distinct industries, including manufacturing, financial services, healthcare, and travel, with Salesforce confirming that no customer data was used during training.
The training scenarios address foundational operational tasks, ranging from generating leads and qualifying pipeline opportunities to resolving multifaceted customer service cases. To build reliable execution, Salesforce applied supervised fine-tuning and reinforcement learning, teaching Koa to navigate sequences of actions and select the precise tools needed to complete them.
Internal evaluations indicate that Koa matches or exceeds leading alternative models on CRM actions within Salesforce’s proprietary CRM Benchmark, while producing three times fewer errors. These benchmark tests measure day-to-day operations such as updating an opportunity, routing an incoming case, and scheduling a follow-up action. Furthermore, Salesforce retains complete control over the model weights, conducting all additional fine-tuning and operational inference directly within its own infrastructure.
“AI is creating a much larger opportunity for software. Every company needs useful AI, tailored to its knowledge, expertise, and work. NVIDIA Nemotron open models give Salesforce the foundation to turn decades of enterprise expertise into specialised AI with Koa, creating a CRM model that can reason and securely take action,” said Jensen Huang, founder and CEO of NVIDIA.
Salesforce is already deploying Koa internally across its own operations and has initiated early customer pilot programmes with organisations including 1-800Accountant, Baxter Credit Union, Engine, Formula 1, UChicago Medicine, and Xero. Broader commercial availability across US regions is scheduled for winter 2026.
Alongside commercial enterprise deployments, Salesforce and NVIDIA are also integrating NVIDIA models and accelerated computing into Missionforce. These post-trained models are designed to support critical public-sector workflows, including procurement, supplier management, and logistics, with flexible deployment options across private clouds and isolated networks.
AWS brings Salesforce data and agents into existing tools
Salesforce data and capabilities can now be accessed directly through Amazon Quick without configuring bespoke custom integrations. Relying on Salesforce’s architecture based on the Model Context Protocol, teams can retrieve pipeline status, account summaries, service cases, and operational analytics straight from within Quick.
AWS is also introducing its suite of AI agents directly into Slack to support workplace collaboration. While the AWS DevOps Agent is available immediately, the AWS Security Agent, AWS FinOps Agent, and AWS Partner Central agents are scheduled to follow in autumn 2026. These embedded assistants are designed to draw upon the ongoing context of Slack conversations when answering queries and collaborating with teams.
“A lot of our engineering work and conversations already happen in Slack. Bringing the AWS DevOps Agent into those conversations allows my teams to spot and resolve issues in their development environments faster, without breaking their flow,” said Muddassar Shaikh, Senior Vice President of Engineering at GoodRx.
The partnership similarly expands foundational AI options, enabling Agentforce customers to access Anthropic and NVIDIA models via Amazon Bedrock, with OpenAI models scheduled to join the roster later. At the data layer, Salesforce Data 360 is gaining Zero Copy access to additional AWS services, including Amazon Aurora, Amazon RDS, and SageMaker Lakehouse. This direct connection allows both cloud environments to utilise enterprise data seamlessly without requiring any duplication.
Looking ahead to autumn 2026, Agentforce Voice and Amazon Connect Customer will introduce direct two-way audio communication between AI agents by leveraging the Agent2Agent protocol. Additionally, Salesforce Clouds are available through the AWS Marketplace across 33 countries, where customers can apply eligible AWS spending commitments directly towards Salesforce solutions, including Agentforce.
Google Cloud connects infrastructure, agents and commerce
Hyperforce is already processing production Salesforce workloads on Google Cloud infrastructure. Salesforce plans to begin migrating selected US enterprise customers onto this setup during Q4 2026, with general availability across North America scheduled for November and expansion into additional global regions planned for 2027.
Salesforce capabilities can also be surfaced directly inside Gemini Enterprise without the need for organisations to engineer custom integrations. Analytical insights from Tableau can be connected using the Model Context Protocol, while Google’s Gemini models remain directly available within Agentforce. This ongoing model availability includes full access through Prompt Builder as well as the Agentforce Reasoning Engine.
“Enterprise AI only pays off when agents can understand a company’s business context and act on it,” said Kevin Ichhpurani, President, Global Partner Ecosystem at Google Cloud. “Gemini Enterprise brings Google’s most advanced AI together with a company’s data, applications and everyday workflows in one platform.”
The commercial relationship extends significantly into online retail. Starting in autumn, shoppers will be able to discover and purchase Commerce Cloud products directly through Google Search, including AI Mode, as well as within the Gemini application. Retail merchants can publish catalogue data to Google Merchant Center, while checkout powered by the Universal Commerce Protocol ensures that payments, regulatory compliance, and order management remain entirely within existing merchant infrastructure.
Data interoperability anchors the technical collaboration. While Salesforce Data 360 and Google BigQuery already support Zero Copy data sharing, the two companies are now expanding that capability into additional regions while adopting both the Iceberg REST Catalog standard and Private Connect.
