The year 2025 was widely tipped by Silicon Valley's elite to be the moment artificial intelligence agents finally joined the workforce. Yet, as the year draws to a close, the promised revolution has stalled, with businesses grappling with trust issues, legacy systems, and a stark lack of measurable returns.
The Unfulfilled Promise of the AI Agent
At the end of 2024, Sam Altman, the chief of OpenAI, forecast that AI agents would completely transform workplaces before 2025 ended. He suggested we might see the first AI agents materially changing company output. Echoing this sentiment, Dario Amodei of Anthropic declared 2025 would be the year AI could match the capabilities of a PhD student.
The confidence seemed well-founded. A survey by IBM of 1,000 developers revealed that 99% were actively building AI agents for business. IBM's Maryam Ashoori confidently stated: "So yes, the answer is that 2025 is going to be the year of the agent." Early signs, like Salesforce CEO Mark Benioff cutting 4,000 customer support roles citing AI agent efficiency, suggested the prediction was on track.
Adoption Lags Behind the Hype
Interest has been undeniably high. A McKinsey survey found 62% of firms experimented with agentic systems, with nearly a quarter scaling them in some business area. In marketing, IT, and knowledge management, agents now handle basic tasks. However, this curiosity has not led to widespread deployment.
In any single business function, no more than 10% of firms are scaling agents. Most companies pushing ahead are doing so in only one or two areas. A significant barrier is the level of trust and organisational readiness required. Unlike simple chatbots, agents need access to core systems and data, raising serious questions about accountability, security, and control.
Structural issues are a major drag. Nearly two-thirds of firms remain in the experimentation or pilot phase for AI overall. Scaling is dominated by large corporations; almost half of firms with revenues above $5bn report enterprise-wide AI programmes, compared to less than a third of smaller businesses.
Legacy Systems and Elusive Returns
Legacy infrastructure is a primary obstacle. Research from Deloitte indicates many organisations lack the data hygiene and governance needed for reliable AI agent operation. These agents must work through existing finance, HR, and order systems, which were never designed for autonomous decision-making.
As a result, performance often disappoints. Gartner found that in marketing technology, while 81% of leaders are piloting or using agents, nearly half say the tools fail to deliver on promised performance. Instead of replacing work, agents frequently create more, necessitating manual overrides, audits, and cleanup when errors occur.
The most significant brake on the predicted boom is the lack of hard returns. While 64% of firms say AI agents enable innovation, only 39% report actual, measurable financial returns. For most, those gains account for under 5% of profits. Cost savings in areas like software engineering are visible but have not translated to substantial bottom-line impact for the majority.
A small group of successful firms, roughly 6% of respondents, take a different approach. They set clear growth goals, redesign strategies with AI in mind from the outset, and invest heavily in talent and governance, committing a larger share of their digital budget to the technology.
For everyone else, AI agents remain in a frustrating middle ground: impressive enough to justify ongoing experiments, but not reliable or valuable enough to earn full autonomy. Consequently, 2025 may be remembered less as the year AI agents took over, and more as the year businesses learned the hard reality of making them work. The technology has arrived, but most organisations have not.
