Full Breakdown
Dreamforce 2026: Enterprise Leaders Say Older AI Models Are Sufficient
By Drooid · · How we work
Business Leaders Cite Cost and Token-Economy Concerns
At the mid-September 2026 Dreamforce conference in the San Francisco Bay Area, senior executives from Salesforce, Anthropic, OpenAI and Nvidia debated the pace of AI development.
Within the Salesforce ecosystem, many attendees argued that the high cost and token-based pricing of frontier models make older, cheaper versions more practical for everyday sales and customer-service tasks. Tim Sanders, chief innovation officer at G2, noted that “the majority of agentic outcomes aren’t driven by frontier capabilities,” emphasizing reliance on last-year models rather than the latest Claude Fable 5.1 or GPT-6 Astra.
Adoption Strategies Favor Established Models
Companies such as Nice, Docusign and Nagarro reported using Salesforce’s Agentforce tools without the newest large-scale models.
Impact on SaaS Margins and Future Outlook
This structural change forces software vendors to rethink pricing and profitability.
Verbatim Quotes
- “It's already hard enough to keep up,” — Alec Bronston, who works at Chicago-based retail data company Spins as senior Salesforce director
- “The frontier models are way ahead already,” — Jaya Rohit Vuyyuru, a vice president at consulting firm SummitX
- “The majority of agentic outcomes aren't driven by frontier capabilities,” — Tim Sanders, chief innovation officer at software reviewing company G2
- “SaaS up to now has had no variable cost to deliver services,” — Tim Sanders, chief innovation officer at software reviewing company G2
- “Astra unambiguously out performs our previous highest-end models (Opus 5, Sol 5.6) on highly complex tasks,” — Patrick Wendell, engineering vice president at Databricks