CIOs prioritize governance and flexibility as AI agents proliferate at Dreamforce

Dreamforce 2026 brought together nearly 44,000 attendees in San Francisco and millions more online, with a major focus on the challenges of scaling AI agents across enterprises. CIOs emphasized the need for robust security, identity, and access controls as AI agents multiply. Key questions emerged, such as whether agents should have the same permissions as human employees and how to maintain human oversight as agent numbers grow from dozens to thousands.

The risks of ungoverned AI agents were highlighted, as mistakes could lead to real business consequences, such as incorrect customer interactions or unauthorized actions. Beyond technical hurdles, CIOs grappled with organizational challenges, including responsibility for agent development, balancing experimentation with governance, and managing costs to avoid overuse of expensive models.

Salesforce introduced the AI Harness, a layer designed to bridge AI models with enterprise systems, providing agents with real business context. This approach ensures that models, while capable of analysis and planning, can act with awareness of specific business needs, such as customer history or logistical issues. The AI Harness integrates with existing IT infrastructure, including data, applications, and governance policies, allowing CIOs to leverage their current systems rather than starting from scratch.

Openness was another key theme, with CIOs seeking flexibility to choose the best AI models for their needs without vendor lock-in. Salesforce and NVIDIA unveiled Koa, a CRM reasoning model tailored for Agentforce, helping agents navigate complex workflows. The event underscored the urgency of managing AI agent proliferation, with CIOs focusing on maintaining control while enabling innovation.