AI Trends for 2026: What Do Entrepreneurs Need to Know?
English Let's eliminate theAI agency, that is, it performs tasks independently: it is a safe bet for the 'hottest AI trend' of the future. Agentic AI seems to be on the inevitable rise: everyone in the world of technology providers and analysts is excited about the prospect of having AI programs that work together to do real work instead of just generating content, even if no one is entirely sure how it will work. Some IT leaders think they already have it (the37%, in an upcoming survey sponsored by UiPath on252US IT leaders); most expect this to happen soon and are ready to invest in this sector (the68%within six months or less); some skeptics (whom we met mainly during interviews) believe that this is mainly an advertising campaign by the suppliers. Most managers in technology-related roles believe that these autonomous and collaborative AI programs will be based primarily on targeted generative AI bots that will perform specific tasks. Most believe there will be a network of these agents, and many hope that agent ecosystems will require less human intervention than AI has required in the past. Some believe the technology will be entirely orchestrated by robotic process automation tools; others propose that agents will be retrieved from enterprise transaction systems; still others hypothesize the emergence of a 'super agent' that will control everything.
Here's what we think: There will be (and in some cases already are) Generative AI (Gen AI) bots that will follow people's orders on specific content creation tasks. You will need more than one of these Gen AI tools to do something meaningful, like make a travel reservation or conduct a banking transaction. But these systems still work by predicting the next word, and sometimes this will lead toerrorsoinaccuracies. So there will still be a need for humans to check them from time to time. The first agents will be those forsmall structured internal taskswhich involve little money, for example helping to change the password from an IT point of view or booking holidays in the HR systems. We don't think it's very likely that companies will use these agents with real customers spending real money in the near term, unless there is the possibility of a human review or cancellation of a transaction. As a result, we don't expect this technology to have a significant impact on the human workforce, except for new jobs that involve writing blog posts about agentic AI (wait, can agents do that?).
Measuring the results of Generative AI experiments
One reason everyone is excited about agents is that it is still difficult to demonstrate the economic value of Gen AI. In the AI trends article from a few years ago we argued that the value of Gen AI had yet to be proven. Data and AI leaders who participated in Randy's “2025 AI & Data Leadership Executive Benchmark Survey” said they are confident that Gen AI is generating value:58%stated that his organization achieved an exponential increase inproductivityo dell’efficiencythanks to AI, presumably especially generative AI. Another16%claimed to have “freed knowledge workers from mundane tasks” through the use of Gen AI tools. Let's hope that these highly positive beliefs are correct.
But companies shouldn't place that trust blindly. Very few companies carefully measure productivity gains or try to understand what knowledge workers freed from routine work do with the time they have. Only a few academic studies have measured Gen AI's productivity gains, and when they have, they have generally found someimprovements, but not exponential. Goldman Sachs is one of the rare firms that has measured productivity gains in the programming industry. Developers reported that their productivity increased by approximately20%. Most similar studies have found contingent factors in productivity, where inexperienced workers get more benefits (as in customer service and consulting) or experienced workers get better results (as in code generation).
In many cases, the best way to measure productivity gains will be to establishcontrolled experiments. For example, a company might ask one group of marketers to use Gen AI to create content without human review, another to use it with human review, and a control group not to use it at all. Again, few companies are doing this, and this will have to change. Given that Gen AI is currently primarily about content generation for many businesses, if we want to truly understand its benefits, we will also need to start measuring the quality of content. This is notoriously difficult to do with the results of intellectual work. However, if Gen AI helps write blog posts much faster, but the posts are boring and imprecise, it is important to measure this: in that particular use case there will be little benefit.
The sad reality is that, if many organizations were to actually achieve exponential increases in productivity, those improvements could be measured inlayoffson a large scale. But there are no signs of mass layoffs in the employment statistics. Furthermore, the winner of the 2025 Nobel Prize in Economics, Daron Acemoglu of MIT, commented that so far we have not seen real increases in productivity thanks to AI and he does not expect anything exceptional in the coming years, perhaps an increase in0,5%over the next decade. Regardless, if companies really want to see and profit from Gen AI, they will need to measure and experience its benefits.
The reality of data-driven culture asserts itself
We seem to be realizing that Gen AI is very interesting. But it doesn't change everything, it specifically changes cultural attributes in the long term. In our 2023 trends article, we noted that Randy's survey found that the percentage of companies surveyed that said they had “created a data- and AI-driven organization” and had “established a data- and AI-driven organizational culture” had doubled from the previous year (since24%al48%for creating organizations driven by data and AI and from21%al43%for establishing data-driven cultures). We were both quite surprised by this drastic improvement and attributed the changes to Gen AI, as it was widely publicized and quickly adopted by organizations.
The numbers have come back a little more down to earth. The37%of respondents said they work in a data- and AI-driven organization, while the33%said it has a data- and AI-driven culture. It is still positive that data and AI leaders believe their organizations have improved in this regard compared to the past, but our long-term prediction is that generative AI alone is not enough to make organizations and cultures data-driven.
In the same survey, the92%of respondents said they believe cultural and change management challenges are the biggest barrier to becoming data- and AI-driven. This suggests that any technology alone is insufficient. It is worth noting that the majority of employees interviewed were from organizationstraditionalfounded more than a generation ago and with a history of transformationgradual. Many of these companies have done more to execute their digital strategies during the pandemic than in the previous two decades.
Unstructured data is important again
Gen AI has had another impact on organizations: it is making unstructured data relevant again. In the 2025 “AI & Data Leadership Executive Benchmark Survey”, the94%of leaders in data and AI said interest in AI is leading to a greater focus on data. Since traditional analytical AI has been around for several decades, we think they were referring to the impact of Gen AI. In another survey we cited in last year's AI trends article, there was substantial evidence that most companies had not yet begun to truly manage data to prepare for Gen AI.
The vast majority of data that Gen AI works with is relatively unstructured, in forms such as text, images, videos and the like. A manager at a large insurance company recently shared with Randy that the97%of the company's data was unstructured. Many companies are interested in using Gen AI to help manage and provide access to their data and documents, typically using an approach called retrieval-augmented generation, or RAG. However, some companies haven't worked much on their unstructured data since the days of knowledge management more than 20 years ago. They focused on structured data, typically rows and columns of numbers from transactional systems.
To declutter unstructured data, organizations must select the best examples of each document type, tag or graph the content, and load it into the system (welcome to the arcane world of embeddings, vector databases, and similarity search algorithms). These approaches offer significant benefits in terms of access to knowledge for employees, which is why many organizations are pursuing them. However, this work still requires a great deal of human effort. At some point, perhaps, we'll be able to simply load tons of internal documents into a Gen AI prompt window. Even when this is possible, there will still be a need for considerablehuman careof the data, because ChatGpt is not able to establish which is the best among 20 different sales proposals.
Who should manage data and AI?
Perhaps it should come as no surprise that while data and attempts to harness it with AI are receiving more and more attention and investment from organizations, the data leadership function itself continues to struggle. The role is still relatively nascent: just the12%of organizations in Randy's first annual survey in 2012 had appointed a chief data officer. Progress is being made: the85%of organizations in Randy's latest survey have appointed a Chief Data Officer, and a growing percentage of these data leaders are primarily focused ongrowth,innovationethe transformation(rather than avoiding risks or regulatory issues). More organizations have also appointed Chief AI officers – surprisingly33%.
As these roles continue to evolve, organizations continue to struggle with their mandates, responsibilities and hierarchical structures. Less than half of data leaders (mostly chief data officers) who responded to Randy's AI & Data Leadership Executive Benchmark Survey said their function is very effective and well-established, and only51%stated that they believe their work is well understood within their organizations. We are not yet sure whether the responsibilities of a Chief AI officer and a Chief data (and analytics-AI) officer (CDAO) require separate roles, although some organizations, including Capital One and Cleveland Clinic, have established the Chief AI officer role as an equal to the Chief data officer.
The only thing we can say with certainty is that theleadership questionin the field of data and AI is destined to grow, regardless of the form, structure and methods that this demand will take. We are undecided about the broader future of data management and AI. Randy strongly believes that the CDAO role should be a corporate role that reports to corporate leadership. He notes that the36%of data and AI leaders in its survey this year reported to the CEO, President or COO. Randy firmly believes that data and AI leaders must deliver measurable business value and understand and speak the language of the business.
Tom agrees that technology leaders need to focus more on business value. However, as we argued in the 2024 Trends Report, he believes there are too many “technology leaders” in most organizations, including CDAOs. Many of these CDAOs find that their internal customers are confused by all the C-level technology managers and that the proliferation of such roles makes collaboration difficult and unlikely to report to the CEO. Tom would prefer to see “supertech leaders,” with all technology roles reporting to them, as is the case in a growing number of companies that have promoted transformation-focused CIOs to fill this role. Whatever the right answer, it's clear that organizations need to step in and make it happenthose who manage the data are respected as much as the data itself.
Translation of the original article “Five Trends in AI and Data Science for 2025”, January 2025