Why AI agent adopters will leave chat users behind

At this point, most companies have implemented AI chat systems. They’re teaching employees how to prompt ChatGPT effectively and measuring modest productivity gains as a result. But a subset of these companies have also deployed AI agents that deliver exponential improvements in productivity, efficiency, and service — making chat-based AI look incremental by comparison. 

This differentiation isn't just technological. It's fundamental to how work gets done.

The productivity ceiling of AI chat

The Box State of AI in the Enterprise report found that 94% of surveyed companies are using AI in some way, but so far, the major wave of AI adoption across most enterprise organizations has centered on basic use cases like chat interfaces. Employees type questions, receive answers, and manually integrate insights into their work.

These organizations are seeing measurable gains from AI chat and basic automation. But the gains quickly hit a hard ceiling. "While most individuals, teams, and organizations have gotten around to implementing AI chat systems, there really are two different worlds of AI adoption right now,” says Box CEO Aaron Levie, “The type-ahead or chat interaction paradigm of AI maxes out at double-digit productivity gains, because you're inherently rate-limited by how fast you can type or interact with the system.

In other words, the bottleneck isn't AI's capability. It's the limit on the speed of human interaction with AI. An employee can only ask so many questions, review so many responses, and manually integrate so many suggestions at a time. In this scenario, the human remains the primary worker, with AI serving as a reactive assistant.

Only organizations leaning into agentic AI are able to break through the productivity ceiling.

The parallel processing advantage

Organizations pulling ahead aren't using faster AI; they're using AI differently. Instead of one-to-one human-AI conversations, they’re deploying multiple agents that can simultaneously operate across different processes. As Levie says, "The AI agent model, where you can run many agents in the background in parallel, actually can deliver multiples in productivity gains.”

Box research validates this: companies achieving greater than 25% productivity improvements with AI are significantly more likely to use advanced agentic systems compared to those seeing minimal gains.

Box COO Jon Herstein interviewed LPL Financial’s Head of AI Advisor Solutions, Steven Chien, on his perspective on AI agents.The nationwide wealth management firm has 29,000 independent advisors and is growing quickly. They talked about how AI has the potential to be a huge productivity multiplier — but it has to work alongside regulatory compliance. As Chien notes, “Just bolting on AI at one end of a process isn't going to fundamentally accelerate that process. Companies need to redesign their processes from the ground up with the idea that AI is available. AI agents offer a way to break out of the gen AI paradox.”

With agentic AI, rather than relying on sequential human review, LPL can run compliance agents alongside content creation. For instance, any advertising LPL puts out has to be reviewed for industry compliance, even if it’s just one independent advisor marketing his business. Traditionally, this was a manual process, where a person would review each piece of content. With AI, that process has  improved, but with basic AI tools, a common challenge is what Chien calls “false positives.” A piece of content flagged for potentially non-compliant verbiage might then stall out in a human queue for review.

More sophisticated AI agents, though, have the intelligence and context to know when a piece of collateral presents a real red flag of when it actually passes compliance muster. This way, it doesn’t hit a roadblock that stops work in its tracks. Now, at LPL, instead of humans managing individual queues, agents process work streams continuously, operating 24/7 across global operations. “Helping to reduce the false positives and focus human attention on the cases that really need it is powerful,” says Chien.

At LPL's recent conference for its fastest-growing advisors, Chien discovered a striking pattern: when asked who was using AI for their marketing, all hands went up. "That's not the case in all the councils that we have," Chien notes, “and we do see a strong correlation between the advisors growing effectively and marketing themselves well, and those who use AI."

Agentic AI across platforms

Parallel processing, in the context of agentic AI, isn’t just about two tasks happening at once. It’s also about AI bridging the gap between different systems that previously would have overly complicated tasks.

Matt Lyteson, CIO of Technology Platform Transformation at IBM, calls this “orchestration” and says, “It is absolutely key to us.” For instance, a sales rep preparing for a conversation with a client might have to research:

• The CRM system for open opportunities
• The ERP system to find out whether there’s a two-way conversation happening
• A client interaction system to assess the health of the relationship
• Etcetera

Now, rather than humans manually gathering that information sequentially, IBM deploys agents across different systems simultaneously. As Lyteson says, "If you want to be able to bring it all together and synthesize it very, very quickly, the only way you can do it is through an orchestration layer.”

The workflow redesign imperative

At this point, the Box State of AI report found, 87% of organizations have begun integrating AI agents, but most remain in basic implementation phases. Advanced autonomous operations are limited to 41% of companies. However grand the promise of agentic AI, in many enterprise organizations, it’s difficult to actually enact. "It’s likely not as easy to adopt as the first paradigm was,” says Levie, “because it requires a change in workflow.”

The obstacle isn't technical — it's organizational. You can’t just take AI and add it on top of your existing applications. You have to bake AI into your core experience to see real productivity gains.

As Chien notes, "A lot of existing workflows at large companies haven't really changed that much. So just bolting AI onto one end of a process isn't going to fundamentally accelerate that process. Companies need to redesign their processes from the ground up with the idea that AI is available."

The compounding advantage

While chat-based AI delivers linear improvements tied to the limits of human interaction, agent-based systems scale exponentially because of their parallel processing capability. But the great AI divide isn't really about technology adoption — it's about reimagining how work gets done entirely. 

Organizations must move beyond superficial AI integration to fundamental workflow reimagination where agents handle routine processing while humans focus on strategy, exceptions, and decisions. The leaders will be the ones that most effectively restructure operations around AI agents working at machine speed rather than human speed. As Levie says, "Those that get there are going to see the future faster, and get more compounding returns, than those that don't."

Companies still stuck on the layer of chat-based AI are quickly missing their window of opportunity — while the AI leaders operate in a fundamentally different productivity paradigm.