The AI ‘Agent’ Hype Is Outpacing Reality
The biggest challenge with the term “artificial intelligence (AI) agent” is the lack of consensus on its meaning.
Typically, it’s used to describe autonomous AI systems capable of performing increasingly complex tasks by interacting with external applications. One of the most talked-about examples in Asia is Manus, which went viral earlier this year.
However, “AI agent” has quickly become a marketing buzzword—applied to everything from automated web-surfing tools to bots that supposedly mirror human decision-making and, according to some, could replace human jobs.
In China, where new agentic tools are popping up weekly, companies have been accused of using the label to ride the AI hype wave.
Amid this confusion, SoftBank founder Masayoshi Son recently declared plans to deploy one billion AI agents within his company by year’s end. At a Tokyo business conference, he described these agents as self-replicating, autonomous systems that can attend meetings, make calls, send emails, and evolve independently—functioning around the clock. He even likened them to Senju Kannon, a Buddhist figure with a thousand arms and eyes, to emphasize their productivity potential.
This kind of grandiose rhetoric has become common among tech leaders in recent months.
On July 18, after OpenAI released new agent-like features in ChatGPT, CEO Sam Altman described the experience as a powerful moment reminiscent of artificial general intelligence (AGI)—a term meant to describe AI that equals or exceeds human intelligence, though its definition remains vague and controversial.
Now, midway through what’s being hailed as the “year of the AI agent,” it’s worth examining how we arrived here—and whether “agent” is even the right term. These tools, after all, aren’t exactly James Bond.
The enthusiasm behind AI agents stems from the belief that this next evolution of AI could finally boost worker productivity—or even replace workers—and generate returns on the massive investments in generative AI.
A recent McKinsey report framed agents as the solution to the “generative AI paradox”: although nearly 80% of companies are using GenAI, just as many report little to no measurable impact on their bottom line. The report urged businesses to redesign workflows around AI agents to enhance agility and uncover new revenue streams. Still, it warned of systemic risks, such as “controlled autonomy,” and cautioned that deploying a billion agents within six months is both technically daunting and potentially dangerous.
In truth, developing AI systems with human-like decision-making is a long road. Cars have existed for over a century, yet fully autonomous vehicles remain uncommon—despite repeated claims that mass adoption is just around the corner. AI follows a similar trajectory.
Agents built on large language models excel at solving text-based problems—like coding or research—and at handling repetitive tasks where oversight is minimal. Some models can now perform multi-step tasks with limited human input.
Yet fully autonomous AI agents in the workplace are still at least five years away. Achieving this will require major advances in tech infrastructure, secure access to workplace tools, and new protocols for managing sensitive data and determining accountability when errors occur.
Until then, it’s premature to equate these systems with deities—or even with human workers. Such comparisons only fuel unrealistic expectations and anxiety about job displacement. It’s more pragmatic to treat AI as a powerful tool—capable of meaningful societal impact, but not imbued with its own will.
Business leaders should shift focus from anthropomorphizing machines to applying AI to solve concrete workplace challenges. Only through practical use can AI agents start making a tangible impact on productivity and business outcomes. And without a shared understanding of what an “AI agent” really is, a billion of them could mean everything—or nothing.
