Not-so-obvious AI predictions for 2025

Industry analyst Mark Beccue shares where he thinks AI is headed in 2025, with an eye to underexplored trends and unexpected developments in the industry.

To no one's surprise, AI kept its foot on the gas pedal in 2024. From hallucinations to worldwide election deepfake anxieties, the technology charged ahead like a teenager with a new driver's license. Along the way, it collected its share of "speeding tickets," like model morphing and other growing pains.

But signs of maturity are also emerging. Pragmatists are gaining control, and operationalized enterprise AI is growing. Generative AI is no longer just a futuristic concept; it's a technology that is on the brink of mass-market adoption. According to research from Informa TechTarget's Enterprise Strategy Group, 30% of respondents already had generative AI use cases in production in 2024.

Looking ahead to 2025, what's next for AI? Here are my predictions for the important developments and trends that might not be on everyone's radar.

Agentic AI will roll out slowly

There is clear focus and excitement around agentic AI, but a few things remain less obvious.

Enterprise Strategy Group's research confirmed that the market is excited but apprehensive: 47% of organizations with generative AI in production or proof-of-concept stages viewed AI agents as valuable productivity tools. However, 35% saw them as experimental technologies, 16% had a limited understanding of their effect and 14% were skeptical of their reliability.

An essential first step is settling on a definition for AI agents. The best definition I've heard so far is that agents must have autonomy, meaning the ability to work without human intervention; perception, or the ability to gather and understand environmental data; decision-making capabilities; and the ability to take actions. 

How many "agents" out there today meet those criteria? Not many yet. At IBM Think 2024, Parul Mishra, vice president of product management for IBM Watsonx Assistants and Business Automations, shared IBM's vision for agentic AI.

"We believe generative AI technology can help scale productivity for more than one app," Mishra said. Agents should go beyond understanding natural language, he said, to manage complex, multistep tasks across systems. These steps might include parsing user intent, connecting to other enterprise applications, retrieving and aggregating information, and finally providing a comprehensive response.

Mishra gave the example of an AI assistant handling the user request "I'd like to see all my meetings this week and send a summary to all of my team." That assistant would need to identify the user, access relevant data, determine which systems and applications to use, and then execute the task seamlessly.

In short, a lot has to happen for AI agents as I define them to work effectively. That level of capability will take time to materialize.

AI models must improve, or generative AI will stall

In industries where most enterprises' service-level agreements demand near-perfect uptime or adherence to rigorous quality management standards such as ISO 9001, the accuracy of AI models is inadequate. Organizations require all sorts of mitigation strategies to ensure that generative AI doesn't hallucinate, propagate bias, leak private data and generally create mayhem.

For enterprises, the downsides of dealing with these new models are becoming clear. In Enterprise Strategy Group's research, 37% of organizations with generative AI in production or proof-of-concept stages reported negative effects from hallucinations. This kind of performance is frankly unacceptable, and the way forward isn't better mitigation strategies -- it's better models.

I expect a trend we saw in 2024 to continue: Model providers publish their training data sources, apply metadata tagging and build models on narrower, higher-quality, cleaner training data sets. Otherwise, generative AI will unquestionably stall because the end result represents too much risk.

AI governance will become a top priority

In 2025, AI governance will capture a significant share of AI budgets. Here, I'm using AI governance as an umbrella term encompassing people, processes, culture change, oversight, model guardrails, mitigation strategies, lifecycle monitoring and more.

Traditional AI might be a bit of an unruly kid, but it's nothing compared to the juvenile delinquent that is generative AI. To manage these risks in 2025, AI governance programs will start to look like other centralized risk management programs, such as cybersecurity. The potential risks AI presents are too costly to be handled any other way.

AI writing assistant tools will experience some market blowback

I'm a huge fan of AI, and there are many high-value use cases. I don't believe AI writing assistant tools are one of them.

Ceding the writing process -- emails, posts, other communications -- to these tools diminishes your value in the AI era. When you outsource the writing process, you are not exercising your mind and bringing critical thinking to your work. And if AI can do your writing, you can be replaced by AI.

I think a growing number of individuals will begin to realize this and ditch the tools. Revenues for companies that provide such applications haven't set the world on fire yet, and I think we're going to see market acceptance remain lukewarm. 

Established leaders will dominate AI thought leadership in 2025

In 2025, AI thought leadership will come not from startups or venture capital-funded disruptors, but rather the pragmatic players with longstanding investments in AI. These include the following, in no particular order:

  • Big tech companies, such as Google, AWS, IBM, Microsoft, Hugging Face and Nvidia.
  • Data-focused businesses, such as Databricks, MongoDB and Snowflake.
  • SaaS giants, such as Salesforce, Adobe and ServiceNow.
  • Infrastructure providers, such as AMD, Dell and HPE.

Mark Beccue is a principal analyst covering AI at Informa TechTarget's Enterprise Strategy Group.

Enterprise Strategy Group is a division of Informa TechTarget. Its analysts have business relationships with technology vendors.

Next Steps

Top AI and machine learning trends for this year

Dig Deeper on AI technologies