We Can’t Afford to Rush the March of AI Agents

(Bloomberg Opinion) — If there was a singular buzzword to emerge at Asia’s largest tech conference last week, it was “agents.”

I jotted it down more than a dozen times from various executive talks and seminars at Taiwan’s Computex. Nvidia Corp. Chief Executive Officer Jensen Huang described them as future “digital employees.” An executive at a semiconductor firm referred to agentic AI as “the next paradigm shift.” I watched countless demo videos featuring bots taking on increasingly complex tasks in users’ work and personal lives — from putting together a marketing presentation to turning off the lights in your child’s bedroom after they’ve fallen asleep.

The industry push behind agents, tools that go beyond chatbots to being able to execute a range of actions on their own, isn’t new. The sector has spent the better part of the last six months promising that this is the year of AI agents. But the hype train seemed to kick into overdrive last week, with a spate of global headlines from Alphabet Inc.’s Google, OpenAI, as well as Tokyo-based Sakana AI and Chinese startup Manus. 

Still, there remain foundational cracks that will hold them back in the near term. And that’s a good thing.

For decades, the idea of giving computer systems too much autonomy was mostly presented as a science fiction-esque scenario that usually doesn’t end well for humans. In the earlier days of the current boom, business leaders focused on how the technology will not replace workers, but rather assist them as copilots. So it’s worth unpacking why the rush for agentic AI has evolved into a full-steam ahead race when concerning patterns continue to emerge.  

The simple answer is that these companies are trying to make money. Nearly every discussion of AI agents at the conference was followed by a thinly veiled pitch to enterprises on optimizing business performance. The pressure is on for tech companies to start proving there is a valid path to profitability following all that investment and infrastructure spending. And it isn’t consumers who will pick up the tab, but enterprises.  

Yet the public’s faith in the technology is falling. A global report from KPMG found the perceived trustworthiness of AI systems fell to 56% in 2024 from 63% in 2022. 

Without building a foundation of trust and a basic framework of accountability, these agents will never be able to replace human workers or take on sophisticated tasks on any kind of scale. The researchers found that organizations can build confidence by investing in responsible governance mechanisms that adhere to international standards.

This also means policymakers must keep pace. Apart from Europe, many regions have been slow to roll out regulations for fear of stifling progress. But there needs to be some guardrails in place.

There are other technical kinks to be worked out as well. A key concern is how to give these tools permission to access various websites that have worked tirelessly to keep non-human users out. Granting a bot entry to anything that requires a password or human authentication to access, such as payment information or sensitive data, while maintaining security has proven difficult. It will require a whole-of-industry rewiring, not to mention the resource strain. The evolution from “simple chatbots to reasoning models to agentic AI will require several orders of magnitude more processing capacity,” research firm IDC said in March.

When I recently tried a demo of the viral Manus AI agent, I was impressed by much of what it was able to do, but it easily got tripped up over tasks that would be very simple for a human — such as filling out an online form to make a restaurant reservation that required my credit card number. At the same time, I didn’t feel comfortable sharing this sort of information with a bot and then letting it run free on the internet.

Giving software programs new autonomy opens the door for an enormous amount of risk. One-too-many viral examples of agents going rogue or messing up will further erode trust and slow widespread adoption. A cautious approach is also imperative as this transition impacts peoples’ livelihoods. 

For this next phase of AI to be successful, it must unfold as a marathon and not a sprint.

More From Bloomberg Opinion:

This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

Catherine Thorbecke is a Bloomberg Opinion columnist covering Asia tech. Previously she was a tech reporter at CNN and ABC News.

More stories like this are available on bloomberg.com/opinion