For technical and software engineering teams, agentic AI is the best productivity tool ever – it boosts productivity by lightyears. For business end-users, however, agentic AI is still a work in progress, with murky results.

That’s the word from Aaron Levie , CEO of Box, who recently joined CXOTalk’s Michael Krigsman in a discussion about the promise and peril of Ai and agentic AI. “We are still in the very early stages of what agentic work looks like in the enterprise and what the rollout looks like,” said Levie. “We have an interesting dynamic, which is sort of a tale of two cities" – agentic coding, which is proving itself to be a game-changer, then there is business-focused agentic AI, which still has uncertain benefits.

The key to leveraging AI in the business is to look beyond mere productivity hacks and view if as “a technology for abundance,” Levie urged. He illustrated this with a thought exercise: Look at a challenging part of your business and ask: What would you do if you had unlimited capacity “for combing through information, for using judgment, for accessing data?”

To date, such capacity has tended to be limited by the number and skills of human workers. Instead of working with spreadsheets and ERP systems data, decision-makers could employ compute power to work through a problem.

For example, if a B2B company could deploy agents to go and comb through its customer base, it would have better insights about the right time to have the right message for its customers. If they could deploy unlimited compute, they could have much deeper views on their customers, and their responsiveness to marketing campaigns. Or, an agent could comb through LinkedIn data to identify potential talent. An agent could comb through marketing and sales results to identify where budgets and resources are being wasted.

At this point, “agents are maybe the most technical solution that has ever been deployed to non-technical people," Levie emphasized. “You’re deploying non-deterministic intelligence into the hands of every knowledge worker.” The danger in this is the agent could "run wild and grab the wrong data and produce the wrong report.”

Software developers may love AI, and business leaders may muse about employing AI to replace more expensive and less-available human labor, but the reality soon hits. Those working closely with agents soon realize that human supervision is essential. There’s always a chance that an agent may “do the wrong thing, or the taste of what the agent delivers is going to be off, or it’s going to introduce a bug,” Levie cautioned.

What is needed now are mechanisms that assure that agents are working within the right context, at the right time, with the right guardrails – especially when less-technical business users are involved. Verification is key.

Processes employed to verify software development cannot be applied to knowledge work. “We’ve seen what the promise of agents looks like in coding. We’ve seen what the promise of kind of chatbots look like in knowledge work. Now the question is: what’s the full promise of agents across knowledge work? I think this will be the defining topic certainly over the next few years within the enterprise.”