As generative AI grows out of its toddler phase this year, it will start to move beyond the pains of the early stages of a new technology and mature into what will ultimately define it for the coming years, when agentic AI dominates the industry and enterprises focus on orchestrating agents.
In the last three years, generative AI has developed from a technology that many vendors and enterprises tacked onto software platforms to one in which most tech vendors now provide an agentic capability or copilot.
The early years of the technology brought a flurry of large language models, and it began to appear as if the bigger the parameter and context window size, the better. The technology shifted from large models to smaller models, with a bit of emphasis on open source, with models from Chinese vendors such as DeepSeek and Alibaba.
Over the last year, the emergence of reasoning and thinking models has highlighted that the market has moved from a focus on models to agents. Juxtaposed with that, there is also a significant new emphasis on the importance of data centers.
Sustained Acceleration and Experimentation
The rapid advancements in generative and agentic AI each year since OpenAI released ChatGPT in November 2022 look to continue in 2026. Many in the AI industry and outside observers are paying close attention to the state of the AI market, whether it is in a financial bubble and whether that bubble will pop anytime soon. This year, enterprises will emphasize using AI agents more, with a clearer purpose within their organizations and geopolitical locations, and clearer usage across verticals such as media and retail.
“We’re going to see 2026 as the year of acceleration,” said Bradley Shimmin, an analyst at Futurum Group. “Not in terms of investment because that already happened, but in terms of optimizing the spend that’s going on for data in serving AI.”
He added that decision-makers will have to decide how to save money and what to cut.
These options come amid several possible outcomes related to the AI boom-bubble question, Shimmin said.
“You have this race toward what will either be a pop of the bubble where everyone’s disappointed, and going ‘oh my gosh, why did we invest in this?” he said. “If that doesn’t get reached first, then we might see some bubble shrinking back to normal … or you may see a continued growth that is built upon actual capability.”
The different options stem from the fact that generative AI technology is still a young technology, Shimmin added.
“We’re still learning, we’re still understanding what this spaceship is,” he said. “We find ourselves piloting without really understanding what the thing does and how it actually works.”
Counting the Cost
With most enterprises still in the experimentation phase and still piloting AI projects, 2026 could be the year where there is a push to “get it right,” said Mark Beccue, an analyst at Omdia, a division of Informa TechTarget.
“This is the year AI continues its methodical journey toward pragmatism,” Beccue said.
As enterprises work with generative and agentic AI, one challenge is understanding the risks and ensuring that organizations can afford to use AI tools to address a particular problem, Beccue added.
One way to reduce the risk of experimentation is to lower the high cost associated with AI technology.
“The pressure is on the industry across the board for driving down costs,” Beccue said. “There’s a major concern that costs have to come down for these things to work.”
One expensive area is data centers.
“We will have a bubble, and things will go down if the cost per use or the cost per doing something doesn’t come down,” Beccue said, referring to the idea that enterprises could end up using AI technology less if the cost of AI products and services remains high.
Better Agentic AI Technology, Data and Shadow Agents
This will also be the year when model makers will release more capable multimodal AI models that better support AI agents and enable more orchestration of AI agents.
“Getting multi-vendor agent ecosystems up and running, that actually cooperate predictably and productively, has turned out to be hard,” said William McKeon-White, an analyst at Forrester Research. He added that as interoperability advances, agent orchestration will become more successful.
“It’s still going to be a growing and learning process,” McKeon-White said. “Not everything is going to be created equal; some platforms will just be unhelpful.”
Organizations will also need to learn how to build data environments that are more secure, with better permission systems and optimized for their intended use so that AI agents or AI tools work as intended, McKeon-White added.
“There’s been a lot of investment organizations are placing into making their data environments better, making them more secure,” he said. Without having a healthier data environment, the improvements AI model makers are making might be limited. Governance will also have to improve to function as a check and balance on models and ensure they work correctly.
Increase in Shadow Agents
In addition, enterprises need to govern the use of AI agents across their organizations and be aware of shadow agents, according to Suja Viswesan, vice president of products at IBM.
Shadow agents are agents that employees use that the organization does not approve.
“Enterprises need to have an inventory of what is happening,” Viswesan said. She added that shadow agents could also include organizations that know when an agent’s lifecycle has ended, with metrics and proper governance in place so they can stop it when a non-sanctioned agent tries to access something it’s not supposed to.
“It’s on the enterprise to make sure that they are doing what they are expected to do so that they can have visibility into it,” she continued. “For all these agents and applications that you are running, we need to know the lifecycle.”
Sovereign AI
Beyond the innovation that will emerge in agentic AI and new models, another outlook for 2026 is the continued growth of sovereign AI.
Sovereign AI is the idea of one nation controlling its own AI technology, such as infrastructure and software, to serve its own interests.
In 2026, sovereign AI will increase in popularity and adoption in the U.K., EU countries, and India, according to Beccue of Omdia.
“That’s going to displace some of these U.S. and China-based vendors,” he said. “It’s not going to displace GPUs and CPUs because they can’t build those quickly.”
He added that the sovereign movement will mainly affect multimodal AI models that are deeply rooted in local languages. For example, having an AI model in a language other than English gives regional vendors a better chance of succeeding. It allows enterprises in those regions access to better models tailored to their language. This will lead to more AI initiatives in production.
In the case of India, the market is large enough to rival those of China and the U.S., Beccue said.
“If the sovereign AI regulations there can boost the Indian ecosystem … it gives them the opportunity to innovate, figure out things, thrive there and then perhaps expand outside of India to challenge AI businesses,” he said.
More Industries Influenced by AI
Aside from sovereign AI, more industries and verticals will continue to be influenced by AI technology in 2026, especially retail.
During the past holiday season, the number of shoppers who visited websites using an AI chatbot before making purchases increased.
This year will continue the trend, said Greg Zakowicz, an e-commerce and retail advisor to Omnisend, a marketing automation platform.
“This is going to be a quick evolution of shoppers more and more toward AI as part of their research and discovery process,” Zakowicz said.
He added that in addition to AI platforms like ChatGPT driving traffic to websites, the next opportunity for large and small brands is on-site driven chatbots like Walmart’s Sparky and Amazon Rufus.
“These chatbots provide a natural handoff from a large AI platform to the on-site experience and can help shoppers refine their search,” he said.
Moreover, it’s likely that by the end of the year, there will be full autonomy for shoppers to go to either ChatGPT or another AI chatbot website like Perplexity and buy whatever they want from where they want, Zakowicz added.
Another industry that will continue to see a change is the media. Some fear AI technology will lead to the replacement of many media jobs. However, AI technology might also continue to help in a specific area of entertainment and media.
“AI will not replace creativity but solve the metadata mess,” said Paul Pastor, founder and chief business officer at Quickplay, a software media company that provides over-the-top video distribution. As a cloud- and AI-agnostic company, Pastor said Quickplay integrates technology from Gemini, Twelvelabs, and AWS into its systems.
While in the media and entertainment industries have been using generative AI to dub and create clips from longer-form content, the next step is how to use these tools together.
“That’s what’s been missing,” Pastor said. “How they all operate together to get the efficiency … [and] how to improve the entire ecosystem.”
The need to piece together the AI tools used for media and entertainment will lead to an increase in connectivity points between recommendation engines, ad-tech, CMS, and analytics under a single layer, according to Pastor.
“[AI] will become the connective tissue that links disparate systems, deriving greater value from current assets and driving greater value to consumers by predicting exactly what a user wants and needs to see next to prevent them from wandering off to another platform or worse yet, hit the cancel button,” he said.
With all these innovations expected in 2026, enterprises might feel overwhelmed by the technology and how to use it most effectively. However, enterprises can hold onto one thing, said Shimmin of Futurum Group.
“The upside to the use of AI has been shown and is absolutely rock solid,” he said. “It is just a matter of time, investment, and effort, not just to maintain that value, but to grow it in directions that we don’t even know that we can do, that we haven’t even thought of yet.”
