Why Ai Agents Failed To Take Over In 2025 - It's 'a Story As Old As Time,' Says Deloitte

ebrublue10/iStock/Getty Images Plus

Follow ZDNET: Add america arsenic a preferred source on Google.

ZDNET's cardinal takeaways

• AI agents did not return disconnected successful 2025. 
• Obstacles see bequest systems and architecture. 
• Companies that succeeded were thoughtful astir implementation.

This past twelvemonth was deemed nan year of AI agents by experts and manufacture leaders alike, pinch a committedness to revolutionize really group activity and summation productivity. However, consultancy Deloitte's caller Tech Trends study recovered that these autonomous AI assistants really grounded to return off, identifying some nan obstacles preventing wide take and ways to flooded them. 

Every year, Deloitte releases its Tech Trends report, which examines nan biggest trends of nan past twelvemonth to thief pass endeavor leaders and workers alike astir what to look retired for successful nan upcoming year. Naturally, nan main taxable of nan 17th report, for illustration that of nan past 2 years' reports, which ZDNET covered successful 2023 and 2024, was AI. However, pinch investments successful AI astatine an all-time precocious this year, nan study offered valuable insights connected how to maximize nan ROI connected agentic strategy. 

Also: While Google and OpenAI conflict for exemplary dominance, Anthropic is softly winning nan endeavor AI race

"Getting beyond nan headlines to really attraction connected nan truthful what, and now what, is nan work that Tech Trends tries to do," said Bill Briggs, CTO astatine Deloitte. "The world's going to proceed to beforehand and evolve, and you can't wait, aliases you will beryllium near behind."

Failure of AI agents to return off 

The emergence of agentic AI exertion had endeavor leaders excited astir nan thought of expanding their workforces and upping productivity pinch AI assistants. As cited successful nan report, Gartner predicted that, by 2028, 15% of day-to-day activity decisions will beryllium made autonomously by agents, up from 0% successful 2024, highlighting nan momentum down nan technology. 

"In a way, AI, particularly 3 years ago, benignant of triggered this activity of excitement, enthusiasm from nan C-suite and nan board... but it was treated arsenic if it was thing abstracted by itself, and nan ceiling was beautiful debased connected nan effect and return from that -- and now agents are having that aforesaid [moment]," said Briggs.

Also: Reinventing your profession for nan AI age? Your method accomplishment isn't your astir valuable asset

Deloitte's 2025 Emerging Technology Trends study, which surveys 500 US Tech Leaders, recovered that 30% of nan surveyed organizations are exploring agentic options, pinch 38% piloting solutions and only 14% having solutions fresh to deploy. The number of organizations actively utilizing nan systems successful accumulation is moreover lower, astatine 11%. 

Some companies are not yet adjacent to deploying nan technology, pinch 42% of nan organizations reporting they are still processing their agentic strategy roadmap and 35% having nary strategy successful spot astatine all. 

The obstacles

This slow complaint of deployment is noteworthy because location is imaginable for agentic AI to optimize business operations. However, galore companies are not successful a position to leverage nan technology. 

"You person to person nan investments successful your halfway systems, endeavor software, bequest systems, SAS, to person services to devour and beryllium capable to really get immoderate benignant of activity done, because, astatine nan extremity of nan day, they're [AI agents] still calling nan aforesaid bid systems, pricing systems, finance systems, HR systems, down nan scenes, and astir organizations haven't spent to person nan hygiene to person them fresh to participate," said Briggs. 

Also: Dreading AI occupation cuts? 5 ways to future-proof your profession - earlier it's excessively late

Obstacles identified by nan study included nan bequest endeavor systems that galore organizations still trust on, which were not designed for agentic AI operations and origin bottlenecks successful accessing systems, hindering nan agents' expertise to transportation retired actions and execute tasks. 

Similarly, nan information architectures of nan information repositories, which provender accusation to nan AI agents, are not organized successful a measurement that enables nan AI agents to devour it. Deloitte cited a 2025 study it conducted that recovered that 48% of organizations identified nan searchability of information arsenic a situation to their AI automation strategy, and 47% cited nan reusability of information arsenic an obstacle. 

Also: Stop utilizing ChatGPT for everything: The AI models I usage for research, coding, and more

Lastly, organizations often neglect to create nan due governance and oversight mechanisms for nan agentic systems to run autonomously, arsenic accepted IT governance doesn't relationship for AI agents' expertise to make their ain decisions. 

"You've sewage this furniture connected top, which is nan orchestration/agent ops. How do we instrument, measure, put controls, and thresholds, truthful if we sewage it right, nan metre wouldn't beryllium spinning retired of control, benignant of for illustration we saw pinch nan early days of unreality adoption," said Briggs. 

Thinking astir nan quality role

Deloitte identified a shape among organizations pinch successful implementations of AI: being thoughtful astir really agents are implemented. Business processes were created to fresh quality needs, not those of AI agents, truthful nan displacement to automation intends rethinking existing business processes. 

Rather than conscionable "layer agents onto existing workflows," Deloitte said successful organizations "redesign processes" to return nan champion advantage of AI's agentic capabilities, leaning into their expertise to tackle a precocious measurement of tasks collaboratively without breaks. 

The quality constituent besides involves ensuring that labor successful nan statement are decently trained. According to nan report, 93% of AI walk still goes to technology, while only 7% goes to changing nan civilization and training, and learning. 

Also: Gemini vs. Copilot: I tested nan AI devices connected 7 mundane tasks, and it wasn't moreover close

Briggs said this disproportionality is "out of whack, because that's nan portion wherever almost everything is going to autumn down." Yet, he said nan deficiency of attraction connected training is simply a "story arsenic aged arsenic time" and a repeated shape seen successful galore of nan tech transformations witnessed during his 30 years successful nan industry. 

Working pinch AI agents will besides raise caller questions, specified arsenic who will negociate these AI agents and teams? What will an HR squad for these agents look like? 

Similarly, Microsoft's 2025 Work Trend Index Annual Report explored nan conception of a Frontier Firm, aliases organizations pinch some AI agents and humans moving successful tandem, and recovered that humans will yet lead teams of AI agents, necessitating HR processes for these assistants. 

"We've sewage to rethink astir of our HR process successful a world wherever we're going to increasingly person group moving pinch algorithms and agents and robots," said Briggs.