How AI experts assess value over hype, and why the best agentic AI is all about results

While tools like OpenAI's Operator garners attention, some experts believe truly independent AI agents are still far off.
Clark Dong of ClearObject sees most current AI agents as being novelties with entertainment value rather than practical tools.
But the tech is still in its early transitional phase, and will ultimately silently blend into everyday workflows for maximum utility.

I believe the likes of OpenAI Operator is more or less a novelty in its current form, it has entertainment value much like watching a dog fetch a beer from the fridge on command. We can do it faster and more accurately, but it's fun to watch.

Clark Dong

Head of Gen-AI Development, ClearObject

The content creator-driven hype train surrounding AI is turning into Aesop's The Boy Who Cried Wolf. When every model and agentic feature release is blown up to be huge news in order to stay ahead in the attention economy, how do we parse through what is entertainment, and what is actually groundbreaking?

Most recently, OpenAI's Operator burst onto the scene with excited fanfare, but experts suggest the reality of truly independent AI agents remains a distant goal. One such skeptic is Clark Dong, Head of Gen-AI Development at ClearObject, an AI solutions provider serving manufacturing and retail sectors. Dong sees agents needing time to develop to be able to integrate into our daily lives, meaning it might not be time to celebrate just yet.

Entertainment value: "I believe the likes of OpenAI Operator is more or less a novelty in its current form, it has entertainment value much like watching a dog fetch a beer from the fridge on command. We can do it faster and more accurately, but it's fun to watch," Dong told us in a recent interview.

Transitional phase: According to Dong, current AI agent technology represents a transitional phase rather than the final destination. "I believe the real utility will happen when the technology fades to the background," he explains. "I expect in the near future we won't be tapping on the phone or typing on a computer, we will be speaking to a device, and the AI tech stack behind it, of which an 'AI Agent' is a part of, will carry out the task in the way we used to do it."

For an AI Agent to be truly useful, it needs to have access to relevant 'states'. States like contexts, preferences, databases, relationships with disparate entities. Many more will come to light as we get past the 'demo' phase and try to do useful work with it.

Clark Dong

Head of Gen-AI Development, Clear Object

Relevant context: Dong emphasizes that for AI agents to become truly useful, they need access to relevant contextual information. "For an AI Agent to be truly useful, it needs to have access to relevant 'states'. States like contexts, preferences, databases, relationships with disparate entities," he notes. "Many more will come to light as we get past the 'demo' phase and try to do useful work with it."

Distrust of AI: Another significant hurdle is the control of downstream systems. "It will also need to be able to control downstream systems to produce work products. This is harder to solve because of our inherent distrust of AI. We can trust deterministic automation, but AI is not deterministic," Dong points out.

Who is interacting with who?: Despite these challenges, Dong sees the evolution of AI interfaces as a positive direction: "Until recently, humans have had to learn to change to interact with machines...finally the machines are smart enough to learn to interact with humans. Speech is the most natural 'user interface.'"