Industry executives and experts share their predictions for 2025. Read them in this 17th annual VMblog.com series exclusive.
By Ramprakash Ramamoorthy, Director of AI Research at Zoho
The year 2024 was a banner one for generative AI. What was once an untested but exciting technology only one year prior was able to grow into its teenage years and start making a difference in the lives of everyday employees. It often operated in the background, supporting tasks like data analysis, grammar, task prioritization, and more-though some companies were able to use it to write code or compose important emails.
More is yet to come, and 2025 has major implications for small businesses.
AI for all
In the early days of generative AI, ChatGPT was considered the gold standard of LLMs, and OpenAI served as its most prominent open source alternative. The field has widened substantially, and now most major vendors have produced a proprietary generative AI model for themselves. Open source LLMs have proliferated, as well, with Meta’s LLaMA, Falcon, and others becoming drivers of accessibility-and the rate of growth continues to rise. Additionally, the barrier to entry has never been lower, as many generative AI solutions include low-code and no-code functionality, easing the learning curve for the less tech savvy employee.
Look for 2025 to be the year generative AI achieves full democratization, starting with how the software is being distributed. Rather than fold it into larger packages of apps, some vendors are starting to offer “AI-as-a-service” platforms, catered to small and medium-sized businesses, that don’t require extensive investments in infrastructure and can support efforts businesses are already making rather than requiring a complete technology overhaul. Expect more vendors to step into the AIaaS game and for some larger vendors to start thinking small, as well, by tailoring AI tools for regional use, including optimization for different languages and cultures.
While the initial call for generative AI adoption was met with some hesitancy, expect 2025 to be when holdouts start softening their stances. Generative AI is now increasingly viewed as a reliable tool akin to search engines for knowledge retrieval and synthesis, saving businesses of all sizes countless hours. While it hasn’t surpassed search engines in market dominance, and likely won’t for some time, there’s a noticeable trend towards blending generative capabilities with traditional search for enhanced user experiences. Look for more casual, routine use cases in 2025 as it becomes table stakes in office suites, development platforms, and CRM solutions.
Regulations take hold
The generative AI trend arrived so suddenly that governments haven’t had the chance to consider proper legislation just yet. However, moving into 2025, it’s becoming clear they can’t wait any longer. The prevalence of deepfakes and the potentially troubling nature of biometric data has forced policymakers to pay attention and take action.
We’re learning more every day, and in 2025, discussions around generative AI legislation will follow the trend and become increasingly targeted. Data privacy should remain a prime target for legislation, but also more of the nitty-gritty inner workings of the technology; for example, expect conversations about algorithmic accountability and how language models can begin eliminating bias and establishing standards in training.
We’ve only begun to see the frameworks for such legislation, still in its very nascent stages, but the call for comprehensive regulations will likely push governments to begin piloting sandboxes to test AI regulations in real-world scenarios, increasing innovation.
Rise of modular AI to support ESG
While vendors have been releasing big, splashy AI products, next year will likely see a wave of focused, task-specific LLMs being introduced as companies begin considering ESG and other sustainability principles after learning just how much energy generative AI can consume. Rather than subvert the use of generative AI entirely, expect these companies to start seeking vendors who calibrate differently sized language models to appropriately sized tasks rather than unleashing the full computing power of generative AI upon each minuscule data point entry or automatically generated email.
In most circumstances, four is a fair number of LLMs upon which to build an AI-powered tech stack. It begins with a narrow model focused entirely on a specific task: fixing grammar, double checking finance calculations, and addressing basic queries. Not only will these LLMs operate more quickly than a cumbersome counterpart, but they also require far less energy -it takes more fuel to run a pickup truck than a Mini Cooper, after all, and one is more efficient in stop-and-go traffic. A Small Language Model (SLM) can take on heavier loads like translations or transcript generation, while a Medium Language Model (MLM) handles the next level up by extracting insights from data and producing basic documents on its own. Everything else falls under an LLM, used sparingly.
It’s not a requirement that companies build their own generative AI software to follow the above model. Some vendors are wising up and inserting these efficiencies into their own products, and it’s only a matter of time before others follow suit, whether committed to ESG or not. Wasted energy is still wasted-and costly.
Expansion of opportunities
Best of all, 2025 is when generative AI finally finds its way into the daily routine of everyday folks, not just those in the technology industry. Virtual assistants will evolve from reactive helpers into proactive life managers who can streamline scheduling and anticipate needs. Smart home systems will now have the ability to precisely optimize power and water use, driving sustainability on a large-scale level. Even something as seemingly insignificant as optimizing daily commutes will start having lasting impacts simply by helping people save their most valuable resource: time.
And, while the media is currently dominated by generative AI use cases-gone-wrong, 2025 is going to be when we start reading stories of the technology doing good in the world. By predicting the severity and impact of natural disasters or accelerating healthcare processes, generative AI can form the foundation for life-saving efforts. With so much in store for 2025, not even generative AI itself could predict it all.
##
ABOUT THE AUTHOR
Ramprakash Ramamoorthy joined Zoho Corp as an intern in 2011 and has since established Zoho Labs, dedicated to AI research. As Head of AI at Zoho, he leads a team of approximately 120 engineers in statistical machine learning, computer vision, and natural language processing. Ramprakash ensures AI integration across Zoho’s products, handling over 5 billion API hits per month. His leadership has been pivotal in transforming AI research into practical applications for customers.
