Emerging AI trends to watch for in 2025

AI agents are reshaping workflows, promising to save time and enhance productivity at work. Microsoft describes agents as "the new apps for an AI-powered world," capable of handling tasks such as balancing financial data, compiling reports, and generating customer leads. Examples include Copilot Studio, which is integrated into Microsoft 365. It allows users to create tailored automation templates or even build custom agents without programming skills, freeing them to focus on strategic tasks, according to Microsoft.

The next leap for AI agents is reasoning—the ability to break tasks into logical steps and execute them autonomously, with models like OpenAI's o1 series and Alibaba's QwQ-32B-Preview as clear standouts in the field. TechCrunch reports that Alibaba's model can solve challenging problems and, according to testing by the company, outperform OpenAI's o1-preview on reasoning benchmarks. These reasoning models are more advanced than earlier generative AI tools, with Alibaba's model designed to fact-check itself and reason step-by-step through complex tasks. However, it struggles with common-sense reasoning and maintaining language consistency.

At an industry summit, Scale AI Founder and CEO Alexandr Wang noted that agents could spark a "ChatGPT moment" in 2025, with widespread adoption similar to the early days of generative AI, according to The Verge. Yet challenges remain. "The internet has shockingly little data of humans carrying [out] actions and documenting their thought processes as they go," Wang remarked.

As AI improves its reasoning skills, scaling challenges persist. Training large models like GPT-4 requires massive energy resources equivalent to powering 5,000 American homes for a year, which is far more than its predecessor. Meanwhile, the supply of high-quality training data might run out by 2028. Innovations like test-time compute, which gives models extra processing power for specific tasks, are emerging as solutions, TechCrunch reported. These improvements could make AI more efficient without relying on costly scaling.

Amid a shifting educational landscape, AI tutors could emerge as a possible way to reshape classrooms. Forty-seventh President-elect Donald Trump's proposal to shift educational oversight back to states raises questions about how schools might adapt to a reduced federal role, NPR reported. This shift creates the potential for a surge in AI-driven tutors, according to senior data scientists at the University of Chicago. Educators might also examine how AI might fill gaps in learning support.

Supporters say AI tutors offer myriad benefits, including personalized instruction tailored to students' unique learning needs and greater accessibility. Platforms like Carnegie Learning and Khan Academy already use AI-driven tutors to deliver personalized academic support. These tools' strengths include addressing student weaknesses and creating interactive and engaging experiences by providing instant feedback and resources.

Experts at the University of Chicago say integrating AI into education is "no longer a matter of 'if' but 'how.'" Still, others caution against overreliance on AI systems. Research from the University of North Florida published in the journal Advances in Engineering Innovation noted that educators should be aware of the ethical considerations surrounding data privacy, potential biases in AI algorithms that might further inequalities, and the "challenge of integrating AI seamlessly into existing educational infrastructures."

A Brookings task force also warned that gains in efficiency from automating tasks like grading or lesson planning should not come at the expense of human-centered instruction. Balancing AI's capabilities with traditional teaching approaches will be necessary to ensure technology enhances, rather than diminishes, learning and development. If implemented responsibly, however, AI could bridge educational gaps for underserved communities.

In a not-so-distant future, your digital self could continue telling your story long after you're gone. This is the promise of digital and AI immortality avatars. Research published in the journal Computers in Human Behavior: Artificial Humans found that virtual reality avatars typically used for social interactions are now expanding to social media and therapeutic contexts. These avatars offer varying levels of human resemblance, significantly influencing engagement and trust. This gives them vast potential, from helping individuals preserve their memories to allowing brands to create immersive, interactive campaigns.

Companies like Sensay are creating digital twins that replicate human knowledge, voice, and image, providing a sort of digital immortality. These avatars provide emotional support and help preserve memories that can comfort grieving family members or those supporting dementia patients, leaving a consistent digital legacy.

On the corporate side, brands are also increasingly integrating AI influencers into their broader digital campaigns to engage consumers in new ways. Virtual influencers are being used to promote everything from cosmetics to cars for brands like BMW, Maybelline, and Olaplex, per The Drum.

AI influencers have the advantage of being customizable; since AI avatars can be tailored to different demographics, they can provide a level of personalization that traditional influencers cannot. Markets like China are leading the way in adopting virtual influencers for live streams and product endorsements, suggesting wider adoption is on its way.

Consumers, however, can often sniff out gimmicks or inauthenticity, which raises concerns about the risk of undermining trust with audiences who may be skeptical of avatars and AI-driven personas. As AI avatars continue to evolve, storytelling, character development, and clear ethical guidelines will be crucial to maintaining authenticity and audience engagement.

Virtual assistants can set reminders and answer questions. Soon, however, they'll be able to sense when you're having a rough day and offer a comforting response. Emotion-aware AI, or affective computing, is perhaps the most "out there" trend on this list. It pushes the boundaries of what we expect from technology.

Unlike traditional AI, these assistants don't just complete tasks; they interpret body language. Emotion-aware AI adjusts its responses based on nonverbal cues like facial expressions and vocal tones, similar to human behavior, according to MIT Media Lab scientist Javier Hernandez.

Research published in the Journal of Business Research found that empathic AI-enabled voice assistants, such as Amazon Alexa, can increase user trust and engagement, especially in contexts like voice commerce. Empathic AI assistants combine practical functions with emotional connections to improve user experiences and build brand loyalty. According to researchers, however, careful design is needed to avoid making people uncomfortable, as overly humanlike empathy could cause unease.

This is already changing how companies understand customer experiences. In advertising, Affectiva uses technology that captures people's real-time reactions to advertisements to help brands see how viewers really feel. Cameras track emotional responses while watching commercials to give companies a deeper look at what makes an ad effective.

This approach goes beyond traditional surveys by showing moment-by-moment emotional changes. This technology is also expanding into other areas like automotive safety where it can help detect driver fatigue or distraction, the Massachusetts Institute of Technology Sloan School of Management reported.

Affective computing technologies also have the potential to enhance public services by creating emotionally intelligent interfaces that adapt their approach to improve user experience, according to a 2024 report from Deloitte on affective computing. Emotion-aware technology is also being introduced in health care, which monitors patients' emotions to provide personalized care.

However, as this technology evolves, ethical considerations around privacy and surveillance, consent, and bias must be at the forefront of people's minds. Erik Brynjolfsson, a professor at MIT Sloan, emphasized the importance of designing inclusively, with diverse training datasets and ethical considerations to ensure privacy and cultural sensitivity.

"The paradigm is not human versus machine—it's really machine augmenting human," Brynjolfsson told MIT. "It's human plus machine."

Story editing by Alizah Salario. Additional editing by Kelly Glass. Copy editing by Paris Close.