Salesforce Unveils AI-Powered Interfaces and Purpose-Trained Models at Dreamforce 2026

The recent Dreamforce 2026 conference showcased several key ideas that are transforming enterprises. One of the main takeaways is the shift from traditional software applications to AI-powered interfaces. Khushwant Singh, SVP of Product Management for Platform, explained that agents don't need a screen; they require access to the underlying capabilities, which can be provided through any AI interface.

This change in approach is made possible by Salesforce's AIforce technology, which unlocks tremendous value and new ways of working. With AIforce, users can tap into the entire platform, pulling out relevant information and functionality to create custom interfaces. This eliminates the need for specialized training and makes the power of the platform accessible to a wider range of users.

Another significant announcement at Dreamforce 2026 was the introduction of purpose-trained models, specifically Koa, which is built on the NVIDIA Nemotron 3 Super architecture. These models are optimized for enterprise tasks and provide rightsized intelligence for every task. In internal pilots, Koa proved to be 14% better than internal benchmarks at retaining context during complex conversations.

Slack was also a major focus of the conference, with Rob Seaman, EVP and GM of Slack, highlighting its role as the front door to the enterprise. The company's new work operating system allows teams to work together with AI in shared channels, making it easier for intelligence to compound in the open.

Finally, Salesforce introduced the concept of the Agentic Enterprise, which requires a harness and model working together to run safely and efficiently. Rohan Kumar outlined six core capabilities that any enterprise harness needs: trusted models, context, agency, actions, data, and governance.