At the end of 2024, just over a year ago today, tech behemoth and OpenAI boss Sam Altman predicted that AI agents would completely upturn the workplace before year end.
AI agents are software programmes which break down any given task into smaller, more manageable steps, allowing them to devise and execute them with little to no human oversight.
“In 2025, we may see the first AI agents ‘join the workforce’, and materially change the output of companies”, Altman said at the time.
Similarly, Anthropic boss Dario Amodei posted ahead of this past year that 2025 was going to be “the year that AI can do what a PhD student is able to do”.
Around the same time, IBM released a survey of 1,000 developers building AI for business, revealing that 99 per cent of them were developing AI agents.
IBM’s Maryam Ashoori said: “So yes, the answer is that 2025 is going to be the year of the agent.”
This year saw Mark Benioff, chief executive of enterprise software giant Salesforce, cut his company’s 4,000 customer support team due to AI agents.
“I was able to rebalance my head count on my support,” he said. “I’ve reduced it from 9,000 heads to about 5,000 because I need less heads.”
Companies have ramped up agentic AI investment to similarly cut costs, with PwC revealing in May that of 300 executives, 88 per cent are set to increase AI spend in the next year to implement the new technology.
But, looking back on the year, were all these agentic promises kept? And did the breathless investments bear fruit?
High curiosity
Interest in the adoption of these agents has been undeniably strong, with 62 per cent of firms surveyed by McKinsey experimenting with agentic systems, and nearly a quarter scaling them somewhere in the business.
In marking, IT and knowledge management, agents are already handling and executing basic, repetitive tasks.
But that curiosity hasn’t yet translated into broad deployment.
In any single business function, no more than 10 per cent of firms say they are scaling agents.
Even among those pushing ahead, most are doing so in just one or two areas.
Part of the reason is that agents demand a level of trust and organisational readiness that many firms simply haven’t built yet.
Unlike chatbots, agents require access to systems or data.
That raises uncomfortable questions about accountability, security and control, particularly when mistakes compound over time.
Adoption lags
There is also a structural lag. Nearly two-thirds of firms remain in the experimentation or pilot phase for AI overall, not just agents.
Scaling has been led disproportionately by the largest companies: almost half of firms with revenues above $5bn (£3.7bn) report enterprise wide AI programs, compared with less than a third of smaller businesses.
Legacy systems are a the primary drag. Deloitte’s research found that many organisations lack the data hygiene and governance needed for AI agents to operate reliably.
Agents can only act through the same finance, HR and order systems already in place, and those systems were rarely designed for autonomous decision-making.
As Gartner found in marketing technology alone, while 81 per cent of leaders are piloting or using agents, nearly half say the tools fail to deliver promised performance.
So rather than replacing work, agents often ended up creating more of it, requiring manual overrides, audits and cleanup if something went wrong.
ROI is yet to be seen
But perhaps the biggest brake on the predicted agent boom of 2025 is that hard returns still remain elusive.
While 64 per cent of firms say AI agents are enabling innovation, only 39 per cent report actual, measurable returns.
And, most of those gains account for under five per cent of profits.
Cost savings and revenue rises have been seen at the use-case level, in software engineering or IT and marketing, but they have yet to add up to significant bottom-line change.
The firms who are seeing meaningful returns, roughly six per cent of respondents, look quite different from their peers.
They set growth and innovation goals and redesign their strategies with AI in mind, rather than piling the tech on legacy systems.
They also invest heavily in talent and governance, as well as committing a far larger share of their digital budgets to the technology.
As for everyone else, agents remain stuck in a middle ground, impressive enough to justify experimentation, but not reliable or valuable enough to earn full autonomy.
In that sense, 2025 may be remembered less as the year AI agents took over, and more as the year businesses learned to make them worked.
The tech has arrived, but the organisations, for the most part, have not.