The Last-Mile Challenge Is Real. Here’s How Enterprise AI Will Overcome It

From vacation planning to homework help, the adoption of AI by consumers has happened at breakneck speed, while enterprises often still struggle to go from a first pilot to broad deployment. 

That last-mile challenge will be familiar to anyone who remembers the rollouts of email, broadband, the cloud, and other technology innovations. It’s the gap between staging impressive demos and making those systems work reliably in real-world operations, as early potential rams into the countless challenges that lie between exciting prototype and adoption at scale. Enterprise AI is just the latest innovation to fall into this chasm. 

But the good news is, by historical standards, companies have the opportunity to cross the last mile faster than ever when it comes to enterprise AI. The ones that can embrace change and comprehend what it takes are the ones that will succeed. 

The Traditional Adoption Curve

In the digital era, transformative technologies have typically followed a five- to 10-year adoption curve. Consumer products often move a little faster — U.S. smartphones, for instance, reached 50% penetration just five years after the iPhone was released. But enterprise infrastructure often moves more slowly. That’s because full adoption requires internal consensus and process redesign. Cloud computing, for example, took roughly a decade to achieve mainstream acceptance after Salesforce pioneered software as a service (SaaS) in 1999. Even then, IT departments struggled to trust their data to servers they didn’t control.

Agentic AI isn’t just a technology implementation challenge; it’s a change-management challenge as well.

Diffusion constraints also include hardware costs, infrastructure requirements, or distribution challenges. The first mainframes, for example, were room-sized monsters requiring raised floors for cable management, industrial air conditioning, and a dedicated staff. While broadband services were available in the 1990s, widespread adoption depended on physical wiring, routers, and carrier availability. Email took several decades to travel from Ray Tomlinson’s 1971 ARPANET experiment to standard business infrastructure in the mid- and late 1990s. 

Consumer AI has rocketed ahead — OpenAI’s ChatGPT reached 100 million personal users within two months of its 2023 launch — but enterprise AI is taking longer to achieve liftoff beyond the prototype stage. That has led to a disconnect.

“AI doesn’t conform to the traditional tech adoption pattern,” said Jim Cavalieri, a 25-year Salesforce veteran and advisor to the chair and CEO of the company, “because it’s spreading with consumer-tech speed while demanding enterprise-level organizational change.”

Just think of all the elements agents need to understand a business, perform consistently, and become key elements of a daily workstream. They need context — from a CRM, service platform, data warehouse, and collaboration tools — so they can all operate from a single source of truth. They need metadata to understand the relationships between these systems, the business logic that governs data flow, and specific role-based permissions that control access. They need to be accessible in the place where work is happening. Furthermore, agentic AI isn’t just a technology implementation challenge; it’s a change-management challenge as well. Agentic AI reshapes how knowledge workers think, create, and make decisions, so it triggers organizational, cultural, and incentive-based resistance.

Challenging as those hurdles may be, they are where true business value lies. “The trapped value is always in latency,” said Geoffrey Moore, author of “Crossing the Chasm” and a Salesforce advisor. “Where in our current operating model is the bottleneck that constrains our overall performance? When we know where the bottleneck is, we know where the trapped value is.” 

Salesforce has been closing that gap ever since it launched Agentforce in 2024. Agentforce is not a separate product but an agentic layer natively integrated across the entire Salesforce product portfolio. Instead of having to stitch together dozens of point solutions or rebuild their entire infrastructure with LLMs, customers get one unified, context-aware AI-powered platform where every app is connected and every agent is grounded in the same truth. 

This essential connective bridge helped the number of Agentforce customers in production increase by 70% in Q3 of this fiscal year. “Reducing the last-mile problem is something Salesforce is really focusing on with its suite of products,” said Cavalieri. 

For years, the pieces have been aligning — ubiquitous cloud computing, sophisticated AI models, platforms rich with customer data. Now the key components have clicked into place.

The last mile is closing.  The future belongs to the companies that can bring together the most powerful models and the deepest context to cross it faster than ever before. 

Read more:

• Three keys to navigating the “last mile” of AI adoption
• 3 Salesforce leaders on why context is king in the agentic era
• In 2025, AI grew up — and learned to play by the rules