Dreamforce 2026: Enterprise AI Moves Out of Pilots — The Practical Challenges Ahead

# What changed at Dreamforce 2026

# AIforce: a headless, platform-level interface

AIforce sits above Customer 360 and Data 360 and is meant to make Salesforce context and permissions available outside traditional application screens. Early surfaces include Claudeforce (bringing Salesforce into Anthropic's Claude), Slackforce (Salesforce context inside Slack), and Agentforce Coworker (an AI teammate within Salesforce). The practical result: users may not need to open a Salesforce app to get relevant data or trigger actions.

Implication: user friction can decrease, but teams must map when and where data surfaces and which permissions travel with the surface.

# Multiple models for different jobs

Dreamforce introduced Koa, a CRM-specific reasoning model, alongside integrations with external models (for example, Claude). The explicit point is that enterprises will run a heterogeneous model environment: general-purpose models for broad tasks and specialized models for domain-specific reasoning.

# Agents are becoming role-specific and longer-lived

Salesforce positioned agents as specialized digital roles across service, IT, HR, commerce, supply chain, and sales. Some agents, like an outbound sales agent called Hunter, operate over days or weeks, retain memory, and pursue ongoing goals rather than single interactions.

Implication: leaders must decide what agents may own end-to-end, what requires human approval, when to intervene, and how to assign accountability if an autonomous sequence produces an unexpected outcome.

# Simpler experience, more complex stack

The user interface will become simpler as AI brings context into existing workflows. Behind that surface is a more complex architecture: multiple models, agent runtimes with different time horizons, data pipelines, policy enforcement, and cross-cloud integrations.

Implication: engineering, security, privacy, and product teams need to collaborate on runtime selection, data access patterns, and operational tooling for observability and incident response.

# Practical questions for leaders right now

• How will we decide which model or reasoning engine handles each task?
• What governance controls will enforce permissions and policy as AI surfaces outside core apps?
• How will we log, audit, and attribute decisions made by agents working across days or weeks?

# Bottom line