Well, it’s that time of the year again for our annual predictions for AI in business.
While most high-tech observers attribute the emergence of generative AI in early 2023 to one of the fastest transformations of enterprise technology landscapes in history, we believe that advancements in 2025 will ignite an even higher-value transformation.
Generative AI has played a critical role in democratizing AI’s value for the masses, delivering incremental business ROI, and sparking the imagination of business leaders everywhere.
We begin 2025 with a foundation for accelerating AI innovation cycles, grounded in the consensus view that this year will mark the rise of agentic AI. In our predictions, we won’t reiterate that shared belief; instead, we will focus on the underlying changes that will drive how agentic AI reshapes the AI marketplace once more.
We foresee a year of transformative advancements tackling various real-world barriers to AI adoption. These advancements will allow businesses to elevate their AI goals from improving productivity and reducing costs to higher-ROI use cases that revolutionize their decision-making processes. We believe this will dominate the industry discourse over the next many years, just as Gen AI did for the past two years.
Our top six predictions for AI in 2025:
- LLMs alone run out of gas to fuel enterprise ROI
- AI reasoning gets democratized for all
- Explainable AI emerges as the currency of innovation
- GenAI wars meet the same fate as the browser wars
- Software developers become data scientists
- AI talent augmentation becomes inevitable
What follows is a review of the state of AI in 2024, the rationale behind each of our predictions, and our recommendations for moving forward in 2025. We welcome your thoughts and suggestions.
Watch the Next Frontiers of AI Podcast
Join Scott Hebner, principal analyst for AI at theCUBE Research, and our industry colleague Tim Sanders, VP of Research Insights at G2, on the Next Frontiers of AI podcast as they debate our predictions for AI.
2024 In Review
Before we rationalize our AI predictions for 2025, let’s first provide some perspective on the state of the AI landscape in business and what transpired in 2024.
After two years of accelerating investment, Gen AI has established a foothold in business. Nearly half of white-collar workers now use generative AI in the workplace to explore new ideas (42%), consolidate information (36%), and automate basic tasks (36%), according to an October 2024 “AI in the Workplace” survey by Gallup. The widespread adoption within businesses is also supported by the Enterprise Technology Research (ETR) tracking surveys, finding that 74% have deployed generative AI use cases. The outcome so far? Several studies, including IDC / Bloomberg, have estimated a 3-point productivity gain among workers. Generative AI so far has delivered progress, but not miracles.
Notably, less than 15% are using AI to make predictions, forecast outcomes, make business-critical decisions, automate workflows, or identify solutions to their problems. The use cases have predominately been focused on productivity and creativity. As a result, quantified ROI has been elusive, as it has been more implicit than explicit. In addition, recent ETRdata also indicates a continual shift in ROI expectations into the future.
In 2024, however, enterprises have increased their machine learning (ML) investments to enhance Gen AI’s productivity value and pursue higher-return use cases. ETR survey data shows a 16-point rise since 2024. More importantly, the spending trajectory remains the highest across all technology categories, with a net score of 57%.
Additionally, as we approach 2025, most businesses foresee more transformative AI use cases that aim to reinvent how they operate, create value, and engage with the world. They intend to develop dynamically adaptable workflows that enhance decision-making and problem-solving by equipping their talent with new “superpowers.”
Technically, this means evolving from AI assistants that help people perform tasks to AI agents capable of helping people achieve goals. With agents at their side, individuals will be better equipped to make informed decisions and solve problems. Agents can also expedite outcomes by acting autonomously when human intervention is not feasible, affordable, or warranted.
We believe that 2025 will herald the emergence of Agentic AI and a widespread shift away from the current fixation on generative AI and LLMs. Nonetheless, this transformation necessitates a reevaluation of the technological landscape, as today’s technologies and skills alone are inadequate to exploit the potential value of agentic AI.
The Agentic AI movement is well underway and forms the context for our 2025 predictions. We believe the changes in 2025 will be game-changing. The race for differentiation has started; some will lead, some will lag, and some will simply fail based on their ability to keep up.
Our six predictions also tell a story. In many ways, they provide a ladder of considerations as businesses start applying AI to solve more business-critical challenges, reinvent their operations, create value, and engage with the world.
Let’s dive into the rational for our 2025 predictions.
#1 – LLMs alone run out of gas to fuel ROI
Over the past three years, large language models (LLMs) have become remarkably capable. They have spurred the broader adoption of AI and laid a vital foundation for future AI applications. However, in 2025, most businesses will recognize that LLMs alone are not sufficient to drive the high-value use cases they envision.
While the number of deployed LLMs has more than doubled to over 70% in 2024, only 24% have implemented LLMs at scale, according to a recent Data & AI Leadership Exchange and DataIQ survey. More significantly, just 18% reported a high measurable value. Additionally, several other surveys have indicated a slowdown in new AI deployments, such as a Rexer Analytics study that found half of projects failed to reach production.
Why? This primarily stems from the limitations inherent in today’s LLMs. While they are highly impressive, these systems are constrained by their correlational designs. They navigate vast lakes of data to identify associations, relationships, and anomalies, which are then utilized to predict or generate outcomes. Essentially, they function based on brute force, processing enormous amounts of data through multiple layers of parameters and transformations to determine the next best token until they can predict a final outcome. This is why you often encounter the terms “billions” and “trillions” in reference to the number of tokens and parameters associated with LLMs. However, this gives rise to a series of real-world limitations because they:
- Are prone to inaccuracies and bias
- Assume a prediction is a judgment
- Can confuse correlation with causation
- Do not understand consequences
- Operate as “black boxes” that cannot explain themselves
This challenge becomes even more evident when LLMs face more ambiguous, goal-oriented prompts, which are central to Agentic AI’s promise of value.
For example, a recent study by Apple found that all state-of-the-art LLMs experienced significant declines in accuracy when faced with different versions of the same problem, altered values within a problem, or seemingly relevant yet ultimately irrelevant information.
Collectively, enterprises are increasingly recognizing that LLMs alone risk becoming incubators for real-world problems and are simply inadequate for generating the high ROI use cases they desire, especially those related to business-critical decisions and outcomes.
As a result, we expect a significant shift toward many enterprises actively developing strategies to enhance and complement LLMs through more advanced model ecosystems and software architectures. Instead of relying solely on LLMs, multi-model AI architectures will emerge as crucial in driving a new generation of agentic AI use cases. In 2024, the consensus view from various studies evaluated by theCUBE Researchsuggests that only 10-20% of enterprises have deployed such architectures at present.
To learn more about the critical importance of an extended AI model ecosystem in building Agentic AI systems, read our recent research note on this topic.
#2 – AI reasoning gets democratized for all
As we approach 2025, pioneering companies are moving beyond Gen AI to implement more sophisticated AI use cases that enhance ROI and transform how their enterprises operate, create value, and engage with the world. According to a study by the Boston Consulting Group, two out of three of these companies plan to utilize advanced AI technologies to deploy AI agents capable of “reasoning.” By 2025, we anticipate that the adoption of AI reasoning will become more widespread, driven by a variety of new tools and platforms that make AI reasoning technologies accessible to a wider audience.
AI reasoning represents a significant advancement for AI in business. It will enable businesses to do more than just make predictions, analyze information, identify patterns or anomalies, and automate tasks. They’ll also be able to explore countless scenarios to understand the consequences of various decisions, solve complex problems, and gain clearer insights into the best pathways for achieving their goals. Additionally, it will allow AI agents to act autonomously when human intervention is not feasible, affordable, or necessary. With the real-time decision intelligence provided by AI reasoning, people will know not only WHAT to do but also HOW to do it and WHY certain actions are preferable to others.
According to a survey by DataIku and Databricks, this promise of value is why seven out of ten individuals are expected to adopt AI reasoning techniques by 2026. This aligns with the 2024 Gartner AI Hype Cycle, which projects that AI reasoning and decision intelligence will become high-impact technologies in the coming years.
The challenge lies in the sheer complexity of implementing trusted AI reasoning. Today’s LLMs and generative AI are incredibly impressive, yet also complex. However, those engaged in implementing AI technologies would tell you that it’s complex on an entirely different level.
Technically, the methods on which AI reasons are its inferencing design — that is, how it progresses from a premise to logical consequences to judgments considered true based on other judgments known to be true. Over the last several years, a range of new AI technologies have emerged that are paving the way for progressive degrees of AI reasoning. Generally speaking, they each include greater capabilities to mimic human reasoning:
- Chain-of-Thought: AI that dissects problems into logical steps that are dynamically assessed and refined to reach a solution for a given question.
- Semantic Reasoning: AI that interprets the underlying context, relationships, and concepts in data to process the knowledge required to complete tasks.
- Causal Reasoning: AI that understands how and why events occur, the consequences of actions, and the causal mechanisms that lead to outcomes.
- Multi-agent Reasoning: AI systems where multiple agents work together by sharing knowledge and coordinating actions to achieve a common goal.
Fortunately, an increasing number of pioneering vendors are making significant strides in simplifying the use of AI reasoning for the masses. So, despite the almost unimaginable complexity behind the scenes, incrementally incorporating reasoning into today’s AI environment will become even easier over time.
These vendors are poised to capitalize on the rapidly growing demand for incorporating AI reasoning, decision intelligence, and explainability into AI agents. This demand is expected to reach a 44% CAGR, leading to a $139 billion expansion in the AI market by 2033.
Our prediction for 2025 is that a widespread industry narrative will emerge about the importance of AI reasoning, along with a significant growth period for the new tools and platforms that make AI reasoning technologies accessible to the masses.
To learn more about the advent of reasoning technologies, we’d recommend the following research notes: (1) OpenAI Advances AI Reasoning and (2) The Causal AI Marketplace.
#3 – AI explainability becomes the currency of innovation
As enterprises continue to explore AI-driven automation and decision-making, a significant roadblock is emerging: a lack of explainability. While generative AI and LLMs have enhanced individual productivity by enabling people to analyze information, create content, make predictions, and automate repetitive tasks, businesses are increasingly hesitant to implement AI in critical decision-making. Therefore, we believe that AI explainability and understandability will become the “currency” of innovation. To “purchase your way” into more innovative, higher-value AI use cases, AI models must improve their ability to explain themselves. The better they explain, the faster the pace of innovation, and the more “currency” will flow.
The reason? Trust. Organizations recognize that without clear, logical reasoning that considers consequences behind AI-generated actions and outcomes, adoption will stagnate—especially in areas where compliance, accountability, and risk management are critical. The productivity-enhancing applications of AI will continue to expand, but in high-value, high-stakes business decisions, enterprises remain hesitant.
Technically, LLMs and predictive models are considered “black boxes” that process countless tokens and parameters to determine the next best token through statistical correlation until they arrive at an answer. They don’t understand what was asked or how they generated an outcome, and therefore cannot provide an explanation. Those who have implemented some degree of explanation are merely PREDICTING what they believe the explanation to be. Prediction is not the same as knowing.
As the McKinsey & Company 2024 State of AI study reported, over 60% see inaccuracies, bias, and the concealment of influential factors in AI outcomes compounded by the lack of explainability as a key inhibitor to using AI in business-critical use cases. ETR (Enterprise Technology Research) also found that half of enterprises are slowing deployments due to regulatory reporting concerns.
The fundamental issue is that today’s AI models, including LLMs, do not inherently explain how or why they arrive at specific decisions. At best, they predict patterns and generate responses based on statistical probabilities, but they lack the ability to articulate their reasoning in a way that humans can interrogate and validate. This presents a significant barrier to trust. As AI systems evolve into more autonomous agentic systems capable of making decisions and taking actions on behalf of businesses, the stakes only get higher. If AI cannot demonstrate an understanding of consequences, causality, and reasoning, organizations will be reluctant to deploy it beyond controlled environments, ultimately slowing investment in more ambitious AI-driven innovations.
However, in 2025, we will witness the emergence of new technologies aimed at enhancing AI explainability. This includes innovations in AI decision intelligence and reasoning, along with deterministic modeling, which will provide users with insights into how AI systems reach their conclusions. This development will make it easier to validate and trust these decisions. The ability of AI to not only generate an answer but also explain how and why it chose that answer—step by step—will become the new standard for enterprise adoption. Companies that prioritize AI transparency will gain a competitive edge, while those that deploy “black box” AI models will face increasing resistance from regulators, customers, and internal stakeholders.
For example, to illustrate how these innovations enhance businesses’ ability to achieve real-world explainability and trust, check out this research note on how Scanbuy and Howso realized tenfold growth in AdTech by utilizing these new technologies.
Ultimately, AI explainability will emerge as the currency of innovation in 2025. Organizations that can trust and verify AI decisions will scale deployments into business-critical use cases, unlocking higher ROI and deeper automation. However, if explainability remains an afterthought, AI adoption will slow, and businesses will struggle to justify expanding their AI investments. Today’s generative AI arms race will evolve into a competition focused on trust, transparency, and understandability. This is why the explainable AI solution market is expected to expand by 40% to nearly $11B by 2026.
The future of AI doesn’t belong to the model that thinks the fastest—it belongs to the model that can explain itself and understands not only what to do, but also how to do it and why it’s the best approach. Our prediction is that 2025 will mark a year of significant progress in AI explainability, driven by the emergence of innovative solutions from pioneering vendors. If this doesn’t happen, we may witness a slowdown in the deployment of AI solutions in mission-critical use cases.
#4 – Gen AI wars meet the same fate as the browser wars
Starting in the late 1990s, the browser wars captivated the tech industry as Netscape, Microsoft, and others fiercely competed to dominate what was seen as the “gateway” to the internet. Today, a similar fervor surrounds generative AI, with companies like OpenAI, Google, Meta, and Anthropic vying to establish their models as the “gateway” to AI. We believe that 2025 will mark the year when enterprises move beyond Gen AI, as its standalone significance begins to diminish.
Within the enterprise, the focus will shift toward creating expansive AI ecosystems that build on a Gen AI service to enable high-value use cases specific to their businesses. This parallels the evolution of the browser from an essential tool to a commoditized gateway, where the emphasis shifted from the browser itself to the unique business value it could facilitate.
At that time, selecting the “right” browser was not merely a question of personal preference; it was a choice that could affect the direction of an entire business.
However, as time went on, it became evident that the browser itself—while essential—was only a base. The true value of the internet didn’t depend on which browser users selected, but rather on the technologies and ecosystems developed on top of it. The browser transformed into a means to an end or a reference point, as the differences between Chrome, Safari, Firefox, and Edge became nuances rather than significant differentiators.
This raises an intriguing question: are we witnessing a similar arc unfold in the Gen AI wars? Today, companies like OpenAI, Google, Meta, and Anthropic are engaged in a fierce battle to define the future of AI. Much like the browser wars, the stakes of making the “right” decision are portrayed as enormous, with debates raging over which AI system delivers the best performance, the most accurate outputs, or the smoothest integration.
We believe it will, as today’s fascination with selecting the right Gen AI and LLM offerings will give way to the realization that the real value lies not in Gen AI services themselves but in the AI models and software ecosystems, they enable. Just as browsers became part of a business’s internet software foundation, Gen AI will become an essential yet commoditized AI service that quietly fades into the infrastructure of the modern enterprise, powering more significant innovations in ways we can’t yet fully predict.
Our prediction is that the Gen AI arms race will not define the future of AI. The true AI competition will turn toward creating robust, unified AI model ecosystems that integrate knowledge, automation, and decision intelligence to accelerate real business results. Companies that recognize this change early—by focusing on building value on these Gen AI platforms—will lead the way in AI innovation and ROI in 2025 and beyond.
#5 – Software developers become data scientists
The boundaries between software development, data science, and AI engineering are quickly dissolving, and we predict that 2025 will mark a pivotal shift where software developers increasingly assume the role of data scientists and AI engineers.
This transformation will unfold on two fronts:
- Skill expansion—as developers upskill in machine learning and data science techniques, largely encouraged and funded by their employers.
- Agentic AI systems for data scientists will empower developers to design and build AI models without requiring deep data science expertise.
A GitHub survey found that 9 out of 10 developers use AI coding tools in their workflows, highlighting the growing integration of AI into software development. However, the real value lies in those able also to play the role of an AI engineer – a capability that only an estimated 2 out of 10 of today’s developers possess. Conversely, the U.S. Bureau of Labor Statistics reports that only 19% of data scientists have formal training in software engineering, limiting their ability to develop software.
This shift is not just an opportunity—it may become a necessity. Software developers who fail to evolve into AI engineers risk stagnation as companies increasingly prioritize AI-driven innovation. As businesses integrate AI into their products and services, those who can develop both software and AI will be in high demand. Additionally, since the demand for software developers, data scientists, and AI engineers is expected to grow in strong double digits over the foreseeable future, many companies are likely to kill three birds with one stone.
Furthermore, the incorporation of data science into software development is becoming increasingly evident. Creating and operating software products now requires vast amounts of data, which must be converted into insights and utilized in AI-driven software.
To bridge this skills gap, educational institutions and training programs increasingly offer curricula that blend computer science with software engineering and machine learning. Beyond skill development, a key driver of this transformation will be agentic AI systems. Rather than requiring deep expertise in data pipelines, statistics, or neural networks, developers will be able to use AI agents to guide them through the process of designing, building, training, and deploying AI models. For example, pioneers such as Moveworks and causaLens offer such agents, reducing reliance on specialized AI teams. This means that developers will no longer have to choose between being a traditional programmer or a data scientist—they will naturally take on elements of both roles as AI agents handle the technical complexities.
Looking ahead to 2025, we anticipate a significant shift in this dynamic. The demand for professionals who can seamlessly integrate software development and data science is expected to rise sharply. DevOps Digestestimates that by 2027, half of software engineering organizations will utilize software engineering intelligence platforms, up from just 5% in 2024. This trend underscores the growing importance of AI and data science competencies in software development.
We are likely entering a new era where software developers and data scientists merge into an increasingly unified profession driven by an industry defined by leadership in AI, intelligent automation, and data-driven decision-making.
#6 – AI talent augmentation becomes inevitable
As AI adoption accelerates, enterprises are encountering an unprecedented AI talent crisis—one that we predict will intensify in 2025. Consequently, companies must fundamentally rethink how they build, acquire, and secure AI expertise to remain competitive. We believe that an increasing number of enterprises in 2025 will adopt AI talent augmentation strategies by flexibly outsourcing skills and sub-projects “on demand” and by leveraging agentic AI systems to enhance existing talent.
The statistics should concern all business leaders with AI aspirations. The demand for AI-skilled professionals continues to outpace supply far, creating a 2.3x gap between job postings and available AI talent. Meanwhile, the number of new AI-skilled entrants into the workforce lags 10x behind new AI job openings. According to a recent Boston Consulting Group study, enterprises are already feeling the strain: 7 in 10 enterprises are struggling to find AI talent. And just as concerning, 40% of AI-skilled employees are considering leaving their current jobs for better opportunities. This expanding skills gap is driving an increase in AI salaries, with some companies paying 47% more to secure top AI talent. As reported by Menlo Partners, some pay 2-3x salary premiums and offer massive equity packages to secure the best of the best.
To bridge this gap, organizations will increasingly rely on flexible, on-demand AI talent solutions, such as outsourcing, contractor augmentation, and AI talent marketplaces. As AI-powered transformation extends beyond IT into business functions like finance, marketing, and customer experience, enterprises will need to quickly scale their AI capabilities without the luxury of hiring full-time AI specialists for every requirement. According to TheCUBE Research, another 200,000+ AI job openings will be added in 2025 in the U.S. alone, making full-time hiring an unrealistic strategy for many enterprises. Instead, businesses will supplement their AI workforce with external talent via project-based AI development, AI-as-a-service models, and specialized AI consulting firms. A few emerging leaders in this space experiencing high demand include BairesDev, Index.dev, and SolutionStream.
However, outsourcing alone won’t resolve the AI talent crisis. By 2025, we will see an increasing dependence on AI agents that serve as virtual specialists, enabling non-experts to build AI models, automate tasks, and enhance decision-making. A marketplace is quickly emerging to offer agents to organizations, empowering more business and IT professionals to utilize AI without needing extensive data science expertise. These AI agents act as collaborative AI teammates, guiding users through data science workflows, automating repetitive tasks, and assisting in developing AI models that previously required specialized knowledge. As AI agentic systems advance, enterprises will broaden their AI talent pool by equipping current employees—rather than relying solely on external hiring—to contribute meaningfully to AI initiatives.
In 2025, AI talent augmentation will no longer be optional—it will be essential. Companies that combine on-demand AI talent with AI-powered agentic systems will navigate the AI skills shortage more effectively, while those that rely solely on traditional hiring will struggle to keep up. Businesses that embrace this shift will benefit from the momentum of AI innovation, while those that hesitate will encounter increasing challenges in securing, retaining, and effectively deploying AI talent.
What To DO and When
In our view, these predictions represent an inevitable reshaping of the AI landscape for enterprises. While a valid argument can be made that many of these predictions should be saved for 2026 or even 2027, we argue that, given the accelerating innovation cycles, the desire to expand AI ROI for transformative impact, and the intense competition for top talent, these predictions will have a significant impact in 2025. We recommend you:
- Set your sights beyond Gen AI and build competency in agentic AI.
- Evaluate the vendors that are democratizing AI explainability and reasoning,
- Create an AI-first strategy for your App Dev teams.
- Aggressively pursue AI talent augmentation strategies.
We recommend checking out the following research:
Thanks for reading. Feedback is always appreciated.
As always, feel free to contact me on LinkedIn.
