Salesforce Deepens Cloud Ties with New AI Model

Salesforce is expanding its AI agent capabilities across major cloud platforms, announcing new integrations with Amazon Web Services and Google Cloud. This move brings its autonomous software closer to the underlying data and daily tools that enterprise teams use. The company also introduced Koa, a specialized reasoning model designed specifically for customer relationship management workflows.

These updates were detailed during the Dreamforce 2026 event. The strategic push links Salesforce’s AI agents more tightly with core cloud infrastructure. By embedding its technology into AWS and Google Cloud environments, Salesforce aims to make AI assistance more accessible within existing digital workspaces without requiring complex custom setups.

Koa model targets specific CRM tasks

Koa is built on NVIDIA’s Nemotron 3 Super architecture. It represents the first reasoning model developed explicitly for CRM work within the Agentforce ecosystem. The system was trained using synthetic data modeled on enterprise knowledge gathered over nearly three decades of real-world deployments. This approach ensures that the model understands complex operational tasks without relying on actual customer data.

The training scenarios cover a wide range of industries, including manufacturing, healthcare, and financial services. Koa is designed to handle multi-step processes such as generating leads, qualifying sales opportunities, and resolving customer service cases. Internal evaluations suggest the model performs comparably to or better than leading alternatives, with significantly fewer errors in routine operations like updating opportunities or routing cases.

Salesforce maintains full control over the model weights, conducting all fine-tuning and inference within its own infrastructure. This allows for a secure and tailored deployment. Early pilot programs are already underway with organizations including Formula 1 and UChicago Medicine, with broader commercial availability expected in the US during winter 2026.

Cloud integrations simplify data access

The new AWS integration allows teams to access Salesforce data directly through Amazon Quick. This eliminates the need for bespoke custom integrations. Using the Model Context Protocol, users can retrieve pipeline status, account summaries, and operational analytics seamlessly within their existing workflows.

Additionally, AWS is introducing its suite of AI agents into Slack to support workplace collaboration. The DevOps Agent is available immediately, while Security and FinOps agents are scheduled for later in 2026. These embedded assistants are designed to draw upon operational data to provide context-aware support.

The expanded partnership with Google Cloud includes robust infrastructure and Gemini Enterprise connectivity. According to GN auto tech/cloud, this collaboration enhances commerce capabilities and provides a solid foundation for enterprise-grade AI. These integrations aim to reduce friction for businesses looking to adopt AI agents without overhauling their current tech stack.

Trade-offs and security considerations

While these integrations offer convenience, they also introduce complexity in managing permissions across multiple cloud environments. Salesforce emphasizes that no customer data was used during the training of Koa, addressing privacy concerns. However, relying on third-party cloud providers for agent deployment requires careful attention to data sovereignty and security protocols.