AI Safety Debate Takes Backseat at Dreamforce 2026
The AI safety debate took center stage at Dreamforce 2026, but attendees seemed more concerned about catching up to existing technology than worrying about its potential risks. The conference in San Francisco saw a shift in tone from last year's obsession with the dangers of rapid AI advancements.
Nvidia CEO Jensen Huang urged frontier labs to 'run as fast as you can,' while others, like Salesforce director Alec Bronston, said it was already hard enough to keep up with current models. Many users are still figuring out their AI budgets and determining which models to use from companies like Anthropic or OpenAI.
Executives at the conference emphasized that older and cheaper AI models are sufficient for everyday sales and customer service work, saying that 'the frontier models are way ahead already.' They cited the example of Agentforce tools from Salesforce, which can respond to customer service questions without relying on cutting-edge models like Claude Fable 5.1 or OpenAI's GPT-6 Astra.
Software companies have to adapt both in their use of technology and in how they deliver services as customers shift to tokens, rather than a subscription basis. Nice's Tim Sanders noted that switching from a typical SaaS model could pull down gross margins from 85% to closer to 45%. However, some companies like Databricks are pushing the boundaries with models like GPT-6 Astra.