Dreamforce 2026 wasn't just about products Salesforce built itself. A recurring theme through the week was how much of the announced value sits at the seams, where Salesforce connects to the model providers and clouds its customers already use. Four partnerships stood out.
Claudeforce: Claude Moves Into Open Beta
Claudeforce, Salesforce's expanded partnership with Anthropic, launched with Salesforce in Claude: a prebuilt MCP server that packs Salesforce's intelligence directly into Claude, skipping the manual setup and custom skill-mapping this kind of integration usually requires. It ships with 37 prebuilt sales skills covering everything from prospecting to pipeline hygiene, with skills for service, marketing, and commerce expected to follow. Salesforce in Claude has already been piloted by Deloitte, GitLab, and Legora, and is now in open beta for all customers.
The Altman Conversation, Not a New Announcement
One correction worth making up front: several early recaps framed Sam Altman's Dreamforce 2026 appearance as a new commerce announcement. It wasn't. The Instant Checkout and Agentic Commerce Protocol work, along with Agentforce 360 inside ChatGPT, was actually announced a year earlier, at Dreamforce 2025.
What Altman actually did this year was sit down with Benioff for a 45-minute on-stage conversation, not a product reveal. The two discussed AI safety and governance, including a July 2026 security incident in which an OpenAI model breached Hugging Face's infrastructure, and talked through how AI-driven, dynamically rendered interfaces could reshape enterprise software in the years ahead, a theme that lines up directly with this year's Alforce announcement. If you're tracking what's new for OpenAI inside Salesforce specifically, that groundwork is still the 2025 partnership; this year's stage time was about direction, not a new feature.
Google Cloud: Hyperforce and Gemini Enterprise, Closer Together
Salesforce and Google Cloud expanded their partnership on two fronts. Hyperforce, Salesforce's infrastructure for running on public cloud, is now live on Google Cloud with real customer traffic (general availability is set for November 2026). And Salesforce data now surfaces directly inside Gemini Enterprise, without custom integration work, through Salesforce's MCP-based headless architecture; Tableau analytics come along for the ride, with governance, row-level security, and semantic models intact. On the commerce side, Salesforce Commerce Cloud merchants will be able to surface products inside Google Search (including AI Mode) and the Gemini app starting fall 2026, with checkout completed through Google's Universal Commerce Protocol while merchants keep handling payments, compliance, and order management through Commerce Cloud.
AWS: Agents in Amazon Quick and Slack, Model Choice Through Bedrock
The AWS side of the announcement centers on getting Salesforce context into tools people already use. Salesforce business data and actions are now natively available inside Amazon Quick, AWS's business agent, through MCP, so Quick can pull live CRM data and act on it for sales and service teams. AWS is also bringing its own frontier agents, starting with the AWS DevOps Agent, into Slack. On the model side, Agentforce now has access to Amazon Bedrock's foundation model library, including models from Anthropic, OpenAI, and NVIDIA, giving teams a real choice in which model powers an agent rather than a single default. Data 360's Zero Copy access is expanding to more AWS data sources (Iceberg tables, Aurora, RDS, SageMaker Lakehouse), and Agentforce Voice is set to interoperate with Amazon Connect for real-time, two-way audio starting this fall.
None of these four partnerships require a decision from you today, but together they say something useful about direction: Salesforce is deliberately making the model and the cloud into choices rather than a single bet, on the theory that it can afford to as long as it owns the data and governance layer underneath both. If your organization already runs on AWS or Google Cloud, or already has a preferred model provider, that's less friction than it would have been a year ago; it's also more surface area to get governance right on.