A Marathon Of Innovation: 4 AI Technologies To Watch In 2025

Venkat Viswanathan is Founder and Chairman of LatentView Analytics, a marketing analytics and decision science company.

Every year, roughly 1.1 million people complete a marathon—26.2 miles of uninterrupted running. Those finishers represent 0.01% of the world’s population.

The history of the marathon dates back to the Greco-Persian Wars in 490 BCE. Legend says that Pheidippides, a Greek messenger, ran 40 kilometers or approximately 25 miles from the plains of Marathon to Athens to announce the Athenians’ victory over the Persians—a battle where the Athenians were significantly outnumbered.

Pheidippides collapsed and died of exhaustion after proudly shouting, “We have won!” The marathon as a modern sporting event, inspired by the Pheidippides legend, was introduced at the first modern Olympic Games in Athens in 1896.

In 1908, the current marathon distance of 26.2 miles or 42.195 kilometers was established at the London Olympics. Reining royals King Edward VII and Queen Alexandra desired for their children to be able to watch both the beginning and the end of the event. Starting at Windsor Castle, the marathon wound through the streets of London to White City Stadium—a venue built for the 1908 Olympics—before finishing in front of the stadium’s Royal Box. The additional distance required to get around the arena and in front of the box added a mile and changed to the marathon’s original 25-mile distance.

So, what does this have to do with AI?

From Proof Of Concept To…

We can think of the first marathon in 1896 as a proof of concept—a confirmation that it was possible to bring structure to an idea or invention.

Today, this is analogous to the last several years in business. Technologists played around with what’s possible using generative AI (GenAI). They tried new things and invested resources to see ideas come to fruition, but they knew that the outcome didn’t have to be perfect.

The innovation we have seen in 2024 is like the 1908 marathon. The initial sporting event was refined with a specific purpose, which meant adjusting the components of the marathons that preceded it. The same is true of GenAI. Early concepts deemed successful were built up and introduced to new audiences—revised from their initial offerings to meet the challenges of a wider swath of users or participants.

But now what? 2025 is the year that GenAI moves from proof of concept to scalable solution, driving notable impact. In other words, this is the year that GenAI will grow from an inaugural race to a world sporting event.

2025 Tech to Watch

According to a study by Harvard Kennedy School professor David J. Deming, American adults have embraced GenAI faster than they did the internet or personal computers—yet another sign of the technology’s potential.

Continuing on the momentum, the following GenAI trends are set to catch fire in 2025 and every executive should be paying attention:

Agentic AI

Agentic AI includes AI systems that are meant to act autonomously in real-world environments and are capable of acting without human interference. In 2024, IoT Analytics found that discussions regarding AI continued to decline in earning calls, but agentic AI climbed significantly in the number of mentions.

According to Forbes contributor Bernard Marr, “[A]gentic AI can make decisions, plan actions, and even learn from its experiences—all in pursuit of objectives set by its human creators.” This technology works best in business settings where problem-solving and planning are required like supply chains, emergency response and advanced robotics. In healthcare settings, AI agents are already triaging patients and managing hospital resources.

Small Language Models (SLMs)

Small language models (SLMs) are leaner versions of their large language model (LLM) counterparts. As companies consider their challenges with LLMs, they will now likely turn to SLMs, which require far less computational power and are optimized for accomplishing highly specific tasks like summarization or classification.

Their advantage is their energy efficiency and power in resource-constrained environments like edge devices.

For businesses concerned about the long-term sustainability of LLMs, SLMs offer a more nimble, energy-conscious solution for simpler tasks. They are also significantly democratizing AI access by offering cost-effective yet powerful solutions for small businesses—allowing them to offer AI-enabled service to their customers and compete with vertical power players.

The most recognizable application of SLM technology is Apple Intelligence—Apple’s newly released personal intelligence system for iPhone, iPad and iMac. The bulk of the technology is housed on each user’s device.

In 2025, we are likely to see the release of similar copycat solutions throughout the market.

Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation (RAG) combines the ease of use of generative models like ChatGPT and the retrieval power of a search engine. RAGs retrieve relevant information from external knowledge bases (like databases or the web) to augment responses generated by other models.

They are most useful in enhancing GenAI with up-to-date, accurate information and often hallucinate less than other LLMs.

RAGs will become a differentiator in 2025 for companies building consumer and AI trust because of their significantly improved accuracy. They will be especially useful in more complicated horizontals like legal and compliance, bringing automation to an area of the business that has largely been untouched by GenAI to date.

Multimodal Intelligence

Multimodal Intelligence is an umbrella term for AI systems capable of processing and integrating multiple forms of input—such as text, images, audio and video—to generate insights. This is useful because instead of relying solely on text, models can intake richer input methods like voice.

While business applications of multimodal intelligence are only starting to be understood, the technology will play a pivotal role in the advancement of technologies like the metaverse and AR/VR environments.

What’s Next

2025 will be a pivotal year for businesses ready to scale their GenAI efforts from proof of concept to solutions delivering value. Like the marathon’s journey to its modern form, this next phase will demand focus, strategy and adaptability.

For executives and organizations willing to embrace these advancements, the finish line is clear: achieving meaningful, scalable impact with GenAI while staying ahead in a rapidly evolving technological landscape. The race has begun—now is the time to set the pace.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?