Enterprises struggle to deploy existing AI as vendors race ahead

The deployment gap

A sharp disconnect emerged at Salesforce's Dreamforce conference this week between the pace of AI development and the speed at which enterprises can actually implement the technology. While Nvidia CEO Jensen Huang urged frontier AI labs to "run as fast as you can" from the main stage, the 50,000 attendees on the conference floor described a starkly different reality, according to CNBC.

Business leaders said their organizations are still working to extract value from AI models that are one or two generations behind the current frontier. Many haven't yet resolved fundamental questions about which models to use or how to budget for them.

"It's already hard enough to keep up," Alec Bronston, senior Salesforce director at retail data company Spins, told CNBC. A slowdown in new model releases would create "a lot of opportunity to just even catch up and get our feet wet."

Why it matters

The gap between AI capability and enterprise adoption reveals a fundamental tension in the technology market. While venture capital and competitive pressure drive rapid model development, most businesses are finding that older, proven models already exceed their operational needs. This suggests the bottleneck in AI value creation isn't model performance—it's implementation, integration, and organizational change management.

Older models prove sufficient

Multiple executives told CNBC that models from a year or two ago handle routine sales and customer service tasks effectively. Kevin Lee, technology chief at cloud contact center vendor Nice, said his company deliberately uses models at least one generation behind the frontier. "In large part, with the models that are out there already today, and even one generation behind, they are highly performant and effective at doing the things that our customers need," Lee said.

Tim Sanders, chief innovation officer at software reviewing company G2, argued that cutting-edge capabilities matter little for practical AI agent deployment. "The majority of agentic outcomes aren't driven by frontier capabilities," Sanders told CNBC. "They're driven by last year's AI."

Jaya Rohit Vuyyuru, a vice president at consulting firm SummitX, said the gap is widening. "The frontier models are way ahead already," Vuyyuru told CNBC. "A lot of the customer base is still getting their feet wet."

Budget questions remain unresolved

Many organizations haven't determined whether to commit to paid offerings from Anthropic or OpenAI or pursue cheaper open-source alternatives, according to attendees. Ram Reddy, technology chief for consumer industries at system integrator Nagarro, said his firm waits roughly three months before integrating newly released models. "We are not one of those first early adopters jumping at it," Reddy told CNBC.

Some companies are taking more aggressive approaches. Databricks deployed OpenAI's GPT-6 Astra across its entire 3,500-person developer workforce this week. Docusign CEO Allan Thygesen said his company uses multiple models simultaneously, routing each request to whichever system delivers optimal value for that specific task.

Salesforce announced a revenue target exceeding $63 billion for its fiscal year ending January 2030 at the conference. CEO Marc Benioff has positioned the company's partnership with Anthropic as central to its AI strategy, with up to 1,000 clients enrolled in Claudeforce, a beta product that connects Salesforce data to Anthropic's Claude chatbot.

These details were first reported by CNBC.