How to define an AI agent—and make sure it’s the real deal
“What does that agent have access to, how was it built?” asks CTO.
Getty Images
“What does that agent have access to, how was it built?” asks CTO.
Getty Images
Are these AI agents the droids you’re looking for?
When it comes to understanding how agents work, Trulioo CTO Hal Lonas believes a Star Wars analogy can be most helpful—by explaining the difference between the famous robots C3P0 and R2D2.
“C3P0 does human cyborg relations and is an interpreter, and that’s not agentic,” Lonas said. “R2D2 was given a mission to deliver plans to Obi-Wan Kenobi, and figured it out—he executed a plan, and when he came up against obstacles, overcame them.”
That kind of “autonomy, execution, flexibility, and changing plans” is part of what makes an agent different from an LLM or software.
“At the high level, I like to use the Star Wars analogy; at the lower level, I like to get into the details of what does that agent have access to, how was it built, what’s the sophistication of what it can do based on its access to AI and models,” Lonas told IT Brew.
Cleanout. Despite such clear definitions, there’s been an increase in “agent-washing,” with some software providers rebranding their products as agentic AI despite a lack of autonomous capabilities; an app with a scripted workflow and lower-level automation might be erroneously called “agentic” in marketing copy.
Overstating or outright misrepresenting AI capabilities can be risky and lead to issues with investors and regulators, according to an April post from Harvard Law’s Forum on Corporate Governance.
“This happens with every large technological change,” Meister said. “AI is just the most recent one, and will probably be the biggest one that we’ll face.”
Gourav Pani, CBO at GoDaddy, told IT Brew that he sees the branding problem—which he hasn’t personally encountered—as part of the usual business churn.
“When there’s a surge of money to be made, you’ll have great service providers and not so great service providers,” Pani said. “The leaders will separate themselves from the ones who are not so great because they’re just better at solving customer problems, and customers will just vote with their feet.”
Be honest. When Lonas is looking for a software product, he told IT Brew, he has certain expectations. Whether or not it’s an agent, how the technology is presented is important—as are the capabilities. Don’t R2D2 my C3PO, in other words.
“They should have some autonomy, that’s what I’m paying for,” Lonas said. “On the other hand, I’m worried about limiting or what are the guardrails for the autonomy, what safeguards are in place to keep it within those rails.”
