Where is AI headed in 2025? Predicting the future is difficult, especially in AI, where developments have rapidly accelerated over the last two years.
The perceptions and plans of business decision makers provide a key signal for AI trends. Our recent survey shows that they are bullish on outcomes, yet nervous about some of the challenges. For instance 92% of respondents state that they will increase usage of AI in the next 5 years, and 3 out of 4 are very or even extremely confident that it will add transformational value, whilst half state they have concerns over AI’s bias and lack of transparency.
Another area for insight are the recent developments and adoption of the technology itself, as well as the underlying research. Generative AI has taken the spotlight the recent years, first with consumer-based AI and OpenAI as a clear leader, but now moving into a market with many more players and a push into enterprise applications where AI is deeply embedded into workflow, processes and interactions.
Which also leads to more requirements for governing AI systems, driven by new regulations as well as corporate ethics. And will non-generative AI return back into the spotlight?
Here are 5 specific trends to watch out for.
Generative AI commoditisation and diversification
Who thought that the development of new generative models and services would slow down in 2024 was wrong. New models are still popping up every week, from OpenAI GPT-4o and OpenAI o1 to Google Gemini 1.5 and a swath of open-source models and this will be no different in 2025.
This increased competition means that there are many contenders for the number 1 position in terms of quality, so in future model providers will have differentiate on other factors such as cost, latency, open-source availability and capabilities such as multi-modality.
Agentic AI
Agent-based systems have been a field of AI research from the start, but practical agent-baed systems were either limited to very specific domains such as electricity demand forecasting or or traffic simulations, or were used for broader purposes but then failed to deliver – anyone remember Clippy, the helpful Microsoft assistant?
This has changed with the advent of generative AI. Agents can use generative AI to interpret highlevel and ambiguous user requests, to understand what the capabiltiies are of various ‘tools’ it can use make sense of data and take action, and to generate plans on how to use these tools to achieve the outcome the user wants.
So with generative AI agent-based systems can finally become a reality, and it is also a way to turn generative AI from a passive service to a active and actionable intelligence, letting generative AI out of its cage, but keeping it on a leash.
Non-generative AI will make a comeback
So is AI all about generative AI in 2025? Actually, all the buzz around generative AI has rekindled interest in the use of other forms AI that are less about right-brain creativity and more about left-brain rational and analytical decision-making and problem solving, such as predictive analytics, real time decisioning and process mining.
Either because of specific interest or because particular problems turn out to be left brain rather than right brain problems. In our survey, 95% of respondents states that the increased prevalence of generative AI is directly responsible for the adoption of other AI tools.
Trustworthy AI
Another indication of AI maturing is an increased interest in trustworthy and responsible application of Ai, driven both by regulation such as the EU AI Act, as well as corporate ethics policies that go well beyond this.
It’s not hypothetical superintelligence or artificial general intelligence (AGI) that enterprises have concerns about, but more the limitations of artificial unintelligence and the improper use of it. The discourse is also shifting from expressing mere good intentions to getting real and into actual ways to operationalize all of this both in capabilities as well as in processes and best practices, such as checking and monitoring bias in automated decisions.
AI-ready workforce
AI trends are not just about technology or even regulation, ultimately AI systems are made and used by people. Whereas historically AI was adopted heavily in specific pockets, it is being rolled out on much wider scale, and bundled with company wide AI literacy and education programs.
And what should not be underestimated is that the first cohorts of new joiners are entering the workforce now who have grown up with AI tools, to either fill up their social feeds or to wrestle themselves through college.
At the end of the day, humans will not be replaced by AI’s, but by humans using AI responsibly.
Peter van der Putten
As Director of AI Lab at Pegasystems and assistant professor, Leiden University. Peter van der Putten focusses on business innovation through trustworthy AI. He advises startups and researches the relationship between AI, humanity, and technology, exploring the future impact of artificial intelligence on society.
