Five Big Ideas Coming Out of Dreamforce 2026

Salesforce introduced AIforce, a platform that exposes its 27-year infrastructure—APIs, metadata, permissions, and governance—to any AI interface, eliminating the need for users to log into Salesforce. This shift allows agents to interact directly with enterprise capabilities, enabling natural-language app building and custom interfaces without specialized training. The move aims to democratize access to Salesforce’s core functions, making them usable by non-technical staff and third-party AI systems alike.

Koa, Salesforce’s first purpose-built CRM reasoning model, was developed in partnership with NVIDIA using the Nemotron 3 Super architecture. It was pretrained on open datasets and fine-tuned with synthetic simulations across 14 industry personas, leveraging Salesforce’s decades of CRM workflow expertise. Internal tests showed Koa outperformed benchmarks by 14% in context retention, 15% in answer relevance, and 11% in tool-call precision, reducing agent errors in complex tasks.

Slack unveiled Slack Code and a multiplayer AI model, enabling teams to collaborate with coding agents in shared channels. This transforms AI from a solitary tool into a collaborative workspace where non-technical staff can review and guide development alongside engineers. The approach aims to accelerate software delivery by making AI-driven coding visible and reusable across the organization, fostering collective problem-solving.

Salesforce also introduced an Enterprise AI Harness to address trust and governance challenges in deploying AI agents at scale. The harness provides six core capabilities—discovery, management, evaluation, cost governance, and enterprise-grade control—wrapped around raw model intelligence. Companies like Fulton Bank demonstrated rapid adoption, highlighting the harness’s role in safely transitioning from AI experimentation to operational deployment.