Generative AI is as trendy as it has ever been.
This year, research into AI was awarded Nobel Prizes, and the largest tech companies in the world pumped AI into as many products as possible. The U.S. government promoted AI as a driver in creating a clean-energy economy and a strategic pillar for federal spending. But what’s next for 2025?
The trend of generative AI in the last few months of 2024 points to a greater push for adoption from tech companies. Meanwhile, the results as to whether AI products and processes see ROI for enterprise software buyers are mixed. While it’s difficult to foresee how AI will continue to shape the tech industry, experts have offered predictions based on current trends.
Respondents to an IEEE study in September rated AI as one of the top three areas of technology that will be most critical in 2025 in 58% of cases. Conversely, nearly all respondents (91%) agree that 2025 will see “a generative AI reckoning” regarding what the technology can or should do. Expectations for generative AI are high, but the success of projects leveraging it remains uncertain.
1. AI agents will be the next buzzword
Based on my research and observations, the use of AI agents will surge in 2025.
AI agents are semi-autonomous generative AI that can chain together or interact with applications to carry out instructions in an unstructured environment. For example, Salesforce uses AI agents to call sales leads. As with generative AI, the definition of an agent’s capabilities is unclear. IBM defines it as an AI that can reason through complex problems, such as OpenAI o1. However, not all products billed as AI agents can reason that way.
Regardless of their capabilities, AI agents and their use cases will likely be at the forefront of generative AI marketing in 2025. AI “agents” could be the next stage of evolution for this year’s AI “copilots.” AI agents could spend time working through multi-stage jobs independently while their human counterpart handles another task.
2. AI will both help and hurt security teams
Both cybersecurity attackers and defenders will continue to take advantage of AI in 2025. 2024 has already seen the proliferation of generative AI security products. These products can write code, detect threats, answer thorny questions, or serve as a “rubber duck” for brainstorming.
But generative AI may present information that is inaccurate. Security professionals may spend as much time double-checking the output as they would if they had performed the work themselves. Failing to review such information can lead to broken code and even more security issues.
“As AI tools like ChatGPT and Google Gemini become deeply integrated into business operations, the risk of accidental data exposure skyrockets with new data privacy challenges,” Jeremy Fuchs, cyber security evangelist at Check Point Software Technologies, said in an email to TechRepublic. “In 2025, organizations must move swiftly to implement strict controls and governance over AI usage, ensuring the benefits of these technologies don’t come at the cost of data privacy and security.”
Generative AI models are susceptible to malicious actors like any other software, particularly via jailbreak attacks.
“AI’s growing role in cyber crime is undeniable,” Fuchs explained. “By 2025, AI will not only enhance the scale of attacks but also their sophistication. Phishing attacks will be harder to detect, with AI continuously learning and adapting.”
Generative AI can make conventional methods of identifying phishing emails — poor grammar or out-of-the-blue messages — obsolete. Disinformation security will become more important as AI-generated videos, audio, and text proliferate. As a result, security teams must adapt to both using and defending against generative AI — just as they have adapted to other significant changes in business technology, such as the large-scale migration to the cloud.
3. Businesses will evaluate whether AI delivers ROI
“The pendulum has swung from ‘new AI innovation at any cost’ to a resounding imperative to prove ROI in board rooms across the world,” Uzi Dvir, global CIO at digital adoption platform company WalkMe, said in an email. “Similarly, employees are asking themselves if it’s worth the time and effort to figure out how to use these new technologies for their specific roles.”
Organizations struggle to determine whether generative AI adds value and to what use cases it can make the most difference. Organizations that adopt AI often face high costs and unclear goals. It can be difficult to quantify the benefits of generative AI use, where those benefits manifest, and what to compare them to.
This challenge is a side effect of the integration of generative AI into many other applications. It makes some decision-makers wonder whether generative AI add-ons truly boost the value of those applications. AI tiers can be costly, and over the next year, more companies are expected to rigorously test — and sometimes discard — the features that don’t deliver results.
Many companies that are incorporating generative AI at a large scale are seeing success. At its Q3 earnings call, Google attributed this result to its AI infrastructure and products such as AI Overviews. However, Meta reported that AI may significantly increase capital expenditures, even as user numbers decline.
SEE: Google Cloud is previewing its sixth generation of the AI accelerator Trillium.
4. AI will make a major impact on scientific research
Along with impacting enterprise productivity, contemporary AI has seen significant movement in science.
Four of 2024’s Nobel Prize winners used AI:
- Demis Hassabis and John Jumper of Google DeepMind won the Nobel Prize for Chemistry for predicting the structure of proteins with AlphaFold2.
- John J. Hopfield and Geoffrey Hinton won the Nobel Prize for Physics for their decades-spanning work developing neural networks.
The White House held a summit on Oct. 31 and Nov. 1 about the use of AI in life sciences, highlighting how AI enables solutions to complex challenges in ways that impact the world. This trend is likely to continue into next year as generative AI models grow and mature.
5. The environmental tools made with AI won’t offset its energy toll
Energy efficiency is another buzzword in AI.
But for every use case in which AI can help predict weather patterns or optimize energy use, there is another story about the environmental cost of building the data centers needed to run generative AI. Such construction requires massive amounts of electricity and water — and rising global temperatures only compound the problem. It’s unlikely an equilibrium will be reached in this large-scale problem.
For businesses, though, expect to see companies touting dubious and genuine claims of energy savings and environmental friendliness around AI. Consider the resource use attached to your organization’s AI strategy.
What are the most popular generative AI products?
The most well-known generative AI products are:
- ChatGPT, the OpenAI chatbot
- Google Gemini
- Microsoft Copilot
- GPT-4, the large language model behind ChatGPT
- DALL-E 3, an image generator
What is the most advanced generative AI?
Various tests have been proposed as potential criteria for determining the most advanced generative AI. Some organizations rate their models on human educational benchmarks, such as the International Mathematics Olympiad or Codeforces competitions.
Other evaluations, such as Measuring Massive Multitask Language Understanding, were explicitly created for generative AI. Google’s Gemini Ultra, China Mobile’s Jiutian, and OpenAI’s GPT-4o sit at the top of the MMLU leaderboard today.