The Definitive Guide to Artificial Intelligence Predictions for 2025
-
By
Tim King
, Executive Editor at Solutions Review - Best Practices,
For our 6th annual Insight Jam LIVE!: Strategies for AI Impact. Solutions Review editors sourced this resource guide of artificial intelligence predictions for 2025 from Insight Jam, its community of enterprise tech and AI builders, implementors, and experts. Join Insight Jam free for exclusive expert insights and much more.
As we enter 2025, artificial intelligence continues to revolutionize industries, driving innovation and reshaping how businesses operate. From foundational advancements in generative AI to cutting-edge applications in automation, decision-making, and customer experience, the AI landscape is evolving at an unprecedented pace.
This curation showcases predictions from leading professionals within our enterprise tech and AI community—builders, implementors, and thought leaders who are pushing the boundaries of what AI can achieve. Their insights provide a glimpse into the trends, breakthroughs, and challenges that will define the future of AI in enterprise applications.
From ethical AI adoption and explainability to the integration of AI with emerging technologies, these expert perspectives offer valuable foresight to help organizations harness AI’s transformative potential. Explore this collection to uncover actionable predictions and strategies from the innovators driving AI into its next chapter in 2025.
Artificial Intelligence Predictions from Experts for 2025
Abhinav Asthana, Postman
Venture capital will pivot to profitability in AI investments
After years of investing heavily in AI startups, VCs in 2025 will shift from growth-at-all-costs to sustainable, profitability-centered models. Major players like OpenAI, Microsoft, and Nvidia have become the cornerstones of AI infrastructure, so funding will increasingly go to startups with strong economic foundations and clear paths to revenue. The days of blank checks for speculative AI projects are over; instead, we’ll see focused investment in companies that show measurable value and can survive in a market dominated by industry giants.
Jeremy Kelway, EDB
Data governance and quality will be the biggest barriers to successful and ethical AI adoption
In 2025, data governance, accuracy, and privacy will emerge as the most significant barriers to effective AI adoption. As organizations look to scale AI, the realization will occur that successful AI outcomes are entirely dependent on trustworthy data. Managing and preparing massive amounts of data, ensuring compliance, and maintaining accuracy will provide complex challenges. Enterprises will need to overcome these hurdles by investing in foundational data platforms that include robust governance controls. As a result, we’ll see a stronger emphasis on data stewardship roles and governance frameworks that align with AI initiatives, as businesses recognize that unreliable data directly impacts AI effectiveness.
Seamless integration of AI and data will redefine core business functions
Moving into 2025, AI capabilities will no longer be siloed as a separate technology, we will start to see these technologies woven seamlessly into core business applications, enhancing traditional business functions and customer experiences. In the next phase of AI, we anticipate a graduation from Proof of Concept and narrowly scoped AI initiatives. Instead, organizations are aiming to incorporate AI into the main architecture of their business platforms, treating data and AI as unified capabilities. This transformation will enable businesses to increase productivity and decision-making power by incorporating AI as a foundational component rather than an add-on.
Charles Ruffino, SoftIron
AI will devour enterprise resources by Q3 2025
That large sucking sound of artificial intelligence isn’t just consuming resources—it’s fundamentally reshaping the technological ecosystem with the subtlety of a black hole consuming everything in its gravitational path. The AI gold rush will continue to devour GPUs, energy, and human capital with an insatiability we’ve never imagined.
But here’s the real strategic inflection point: we’re moving beyond mere technology adoption into a more nuanced terrain of AI value stratification. Think of it like a sophisticated restaurant menu, where not every dish is worth the price. Top-tier engineers won’t just be chasing money—they’ll be hunting for meaningful implementation strategies that transform potential into tangible organizational capability.
The AI hype cycle remains a tension-filled boxing match between promise and performance. We predict and demand the emergence of what we’re calling the “AI Value Amplification Index”—a ruthlessly pragmatic framework for measuring genuine technological impact versus marketing hyperbole. Organizations that develop this discernment will separate themselves from those drowning in algorithmic snake oil.
Critical milestones to watch:
- Granular ROI measurements that go beyond surface-level efficiency
- Emergence of specialized AI implementation consulting
- Increasing board-level scrutiny of AI investment strategies
The vortex will have casualties, yes. Entire IT subsectors will be disrupted, budgets will be cannibalized, and more than a few tech executives will discover their AI strategy is more PowerPoint than power tool. But for those who navigate this landscape with surgical precision, the rewards will be transformative.
Gary Orenstein, Bitwarden
AI-enhanced social engineering scams will continue to dominate the threat landscape
The 2024 Bitwarden Cybersecurity Pulse survey found that 89% of tech leaders are already concerned about existing and emerging social engineering tactics enhanced by generative AI, underscoring the heightened risks. In 2025 people will likely adapt to more believable attacks, but the speed and sophistication of these threats may outpace defense measures. The best way to combat these threats will be layered security—combining passwordless solutions, multi-factor authentication (MFA), and continuous education for employees on identifying potential scams.
Joe Regensburger, Immuta
Greater AI reasoning capabilities will broaden the types of business problems that can be solved using LLMs
OpenAI’s GPT-o1 is being developed to better reason through complex tasks and solve more challenging problems. As these types of LLMs gain traction, the real world problems they are tasked to solve require more than just language modeling. These problems require reasoning and inference. Existing LLMs struggle with these types of problems.
Small Language Models (SLMs) will take off as a means of solving more targeted problems with greater cost-efficiency
We need to be more discriminating in what problems we ask LLMs to solve. Many natural language processing (NLP) applications can be solved using more cost efficient models such as GPT-4o-mini, Gemini-flash, etc. Using more cost efficient models means lowering the cost-covering point for the use of LLM services.
Model security—specifically data security, data lifecycle management, and data telemetry—will be a top priority as Commercial-off-the-shelf (COTS) foundational models drive quicker adoption of Generative AI functionality across multiple industries
Enterprises can now build applications around COTS AI models, reducing the need to acquire and maintain specialized hardware, and affording Generative AI companies the opportunity to amortize astronomical training costs across multiple users. This has been a revolution in machine learning, but it carries a cost to security. The fact that there are a relatively small number of models serving a broad number of users, makes these foundational models tempting targets for adversaries both in terms of training and avoidance. We are applying generative AI to more tasks, and empowering generative AI with a degree of autonomy. This increases the responsibility for AI developers to demonstrate that the data they use to train and refine model predictions is clean, timely, and has provable lineage. We will see a greater need for tools which automate the track data usage throughout its lifecycle.
Jeff Elton, ConcertAI
Oncology-specific AI and LLM systems will work together across the entire lifecycle – from discovery to clinical trials
LLMs specific to oncology will allow for the consideration of Agents as “Interpretation Experts” with performance comparable to the highest-trained humans for 90-plus percent of patient decisions. There will be a new generation of multi-modal infrastructure with persistent data analysis occurring over the life of the clinical trial. Trial designs, patient matching, data collection, and real-time analysis will have AI enablement throughout the process.
AI regulations in healthcare will be marked by highly differentiated approaches and AI adoption
AI regulation currently encourages responsible innovation and self-regulation, allowing space for new advancements. Current Gen 1 solutions, often single-LLM-based, are expected to be short-lived and evolve into multi-model, highly tuned systems with domain-specific models and advanced prompt engineering. Most healthcare AI will be run on data locally, edge-deployed, or done within secure, segregated clouds to ensure control, prevent misuse, and protect patient health data. Lastly, leading AI SaaS solutions will heavily publish performance metrics, certify against model drift, and provide transparent data flow and model disclosures. This would be the equivalent of certifying drug safety and manufacturing standards.
Accuracy will be the turning point of AI, supported by LLMs over ambient AI
There will continue to be an increase in the integration of AI in daily workflows and decision-making as AI increases in accuracy and efficiency. 2025/2026 will see the enormous potential of AI as a ‘decision augmentation’ of expert humans. This will come from context-sensitive solutions, LLMs, that can align other LLMs to collect, analyze, and recommend options to clinical teams that are aligned to that specific decision and the unique characteristics of that patient. This needs to and will happen as there are not enough staff and specialists to provide the needed care.
AI will enhance life sciences through advanced Digital Twins and AI-designed drugs
AI-designed drugs will advance into clinical development. Advanced Digital Twins will simulate early-phase clinical trials and will help identify the most beneficial and likely successful drugs in trials. Fully integrated patient-to-trial matching in provider workflows will cut time and costs for late-stage trials by 30 percent. All of this will lower costs and increase the success rates of pharmaceutical R&D processes.
Andy Boyd, Appfire
Market refinement will feel like an AI “bubble burst” but reveal opportunities for business leaders
The SaaS market will evolve through AI-driven innovation and compliance
Looking toward the year ahead, the SaaS industry will be defined by three critical imperatives: adopting AI, upskilling in AI, and navigating increasing regulatory complexity. To stay relevant in the future, SaaS providers must leverage AI to both build great products and also a foundational method to how teams work — to deliver smarter, differentiated products and services. Additionally, Product leaders will need to revisit their organizational structures, ensuring they have the right expertise—individuals who understand how to use and apply AI—to meet the demands of this evolving landscape.
At the same time, organizations must be aware of the changing risk and regulatory landscape. Evolving compliance and cloud regulations will demand a proactive approach to managing risks, including areas of data security, AI regulation and intellectual property concerns. Ultimately, success in 2025 and beyond will hinge on the ability to align technology and people investments with shifting market realities, all while maintaining a proactive approach toward the risk landscape.
Nikolaos Vasiloglou, RelationalAI
From on-prem to cloud-native AI, towards zero-cost token generation
In the early days of Generative AI (GenAI), there were significant concerns about privacy and data leaks. Companies pushed towards on-premise hosting of language models. Given the shortage of GPU supply that needed to catch up significantly with the demand, the cost of hosting and operating LLMs made intelligent application development difficult and expensive. On the other hand, LLM-as-a-service companies not only improved the inference times and throughput but also got into a race to the bottom for the price of token generation.
At the same time, data cloud providers like Snowflake have invested in building their GenAI stack and providing security and privacy guarantees. In 2025, the cost reduction will push the development of simple LLM workloads, such as entity linking into production.
Enterprise agents devour vast volumes of text
Riding on the wave of cheap token generation, enterprises have terabytes of text waiting to be mined to drive better decisions. In the first big data wave that rose in early 2010, companies started mining volumes of generated untouched stored data once the hardware became cheap and ML tools were developed. The conditions for GenAI seem ripe now, so communication data, such as email, Zoom transcripts, Slack messages, Jira tickets, etc., will be consumed massively by agents in the new year that can provide analytics insights and decision support.
Imagine a CRO in an organization with hundreds of complex sales trying to get the accounts’ status and progress. The daily standup meetings were different leads reporting details of each project will be replaced by agents providing dashboards, charts, and alerts with actionable items.
More symbolic knowledge generation
Knowledge Graphs (KG) are the backbone of modern enterprise efficiency. However, for many years, building one was expensive. Language models have been proven to be an excellent assistant for building KGs. Human supervision is still required. The biggest problem has been the motivation for starting the process. Companies can only afford to make a knowledge graph by tying it to an application. Usually, a significant upfront effort is required to build a high-quality, clean version of a KG to start driving an application. GraphRAG is a popular application that can work with an inexact version of a KG and simultaneously deliver value.
GraphRAG quickly provides a KG that companies can use to iterate and perfect over time. As mentioned in the previous section 2025, agents will process massive volumes of textual information and convert unstructured text to symbolic facts as part of the knowledge graph.
The dawn of fine-tuning
While approaching the limits of in-context learning, academia and industry are exploring the value of fine-tuning more. While question answering is well handled with in-context methods like RAG and its variants, there are cases where latency and speed matter, so fine-tuning smaller models makes more sense. Also, we saw LLMs able to solve complicated reasoning problems that can be toys for the moment, like playing chess, solving sudokus, and other puzzles.
There are a lot of enterprise applications, such as planning supply chain optimization, that are based on the same principles. While we expect to see a small adoption in the new year, more exploration and interest will shift toward this paradigm. While there seem to be many lower-hanging fruits, we shouldn’t exclude the possibility of an explosion of use and adoption of LLM applications like this, given the availability of a currently idle tech workforce.
Skip Levens, Quantum
The AI hangover is settling in – Living (and working) with the reality of AI
AI may have grabbed the headlines in 2024, but in 2025 organizations are going to get real about how they want to use AI—and the realities of implementing it. AI today can perform some impressive feats—generate artistic images, answer open-ended questions—actions formerly the province of humans alone. But it can also do a lot of the more ‘boring,’ tedious manual tasks that bog our day-to-day work down.
In 2025, organizations are going to end their AI exploration phase to instead take a deep, realistic look at their need for the technology and how it will meaningfully help their business and customers. And they’ll find that their best minds will not be replaced by AI, but will see how well AI can amplify their expertise. While AI doesn’t create the idea, AI can help make the idea a reality faster. We’re going to start seeing businesses tapping virtual agents and copilots for the tedious work while letting humans do what they do best—be creative.
We’re talking the data race v. the arms race
In the last year, there has been a frenzy around AI, with investors and organizations throwing cash at the buzzy technology. But the real winners are those who saw past the “buzz” and focused on actionable takeaways and what will actually help their organization. We’re finding now that the gold rush isn’t the technology itself, it’s the data that feeds AI and the value it presents. In 2025, organizations that take a more pragmatic approach to AI—and its underlying data infrastructure—will be best prepared to fuel new insights and power discovery.
Those who are leading the data race are the ones who are not only leveraging every scrap of their collected data for differentiated AI outcomes, but those who have an infrastructure and process in place for effectively doing so—managing, organizing, indexing, and cataloging every piece of it. They’ll produce more, faster, and better results than their competitors. In 2025, we’ll start to see who leaps ahead in this new ‘data and algorithm arms race.’
Growing up in the age of AI – What’s real, what’s not?
What happens when almost every piece of ‘born digital’ media seen on the web and social media meets an avalanche of readily available generative AI tools? It means almost everything you see in your digital day could have been generated by AI—and inherently untrustworthy. The effects of this today might provoke a laugh or a gasp for a relatively crude implementation (why do AI images always have the wrong number of fingers?)—but the implications of pervasive and increasingly higher quality gen AI tools will be far reaching. Every business, every walk of life, every institution will need to evaluate their communication strategy, transparency in using these tools, sources of their training data, and more as the technology matures.
Mark Cusack, Yellowbrick Data
Rise of Private LLM Deployments
As concerns over data privacy, cost, and control continue to grow, more companies will opt to deploy private Large Language Models (LLMs) in-house. Companies will prioritize data privacy by avoiding the sharing of sensitive information with third-party models like OpenAI, ensuring their data is not used to train competitors. In addition, the unpredictability of cloud costs will push businesses to run models internally. The increasing availability and decreasing cost of commodity GPUs will further make it more affordable for companies to manage LLMs on-premises rather than relying on cloud providers. These private LLM deployments will give businesses greater control over both their data and costs, positioning on-premises solutions as the preferred choice for many.
Tim Eades, Anetac
The AI Threat: It’s Real, and It’s Here
We’re at a defining moment in cybersecurity that will determine organizational survival. Transform or be transformed by a competitor—this isn’t a slogan, it’s a survival mandate. As organizations integrate AI into their business and security operations, they face increased identity vulnerabilities. This requires enhancing organizational visibility within networks. AI amplifies cyber threats exponentially: it makes good hackers great and great hackers scale. Organizations that fail to implement comprehensive monitoring mechanisms will face devastating attacks. It’s not a question of if, but when.
We’re seeing the first wave of attacks, and they’re already mind-blowing. Take the Wiz CEO incident—where attackers used AI to perfectly replicate an executive’s voice to authorize a fraudulent transfer, bypassing traditional security measures. This represents just the first inning of AI-enhanced cyber attacks and phishing attempts. Without robust visibility solutions that enable real-time detection of anomalies—such as unusual route updates, unexpected configuration changes, or suspicious account activities—organizations remain critically vulnerable.
Drawing from collaborative guidance by top security agencies like the CISA, NSA, and FBI, critical infrastructure and organizations across the globe must prioritize enhanced visibility and cybersecurity hardening. As AI enables cyber adversaries to scale their operations, expect nation-state actors to increasingly target critical infrastructure and organizations essential to modern life—disrupting healthcare, supply chains, and financial services.
Molly Presley, Hammerspace
GPU Demand Soars, but AI Adoption has Companies Rethink Resource Allocation
As we enter 2025, the AI industry faces an unexpected situation: there’s a huge demand for GPUs worldwide, yet many of these powerful chips aren’t being fully used. While companies invested heavily in GPU-based infrastructure, many continue to struggle to apply these chips to AI workloads, instead redirecting them toward non-AI applications. The expected AI-driven boom remains slower than anticipated.
We will continue to see companies be more selective with GPU allocations, as companies focus on areas where the impact of AI in areas like data analytics and cloud computing enhancements – rather than emerging AI initiatives. Additionally, as developers become more resource-conscious, the focus on optimizing algorithms for available hardware, leveraging CPU-bound AI, and adopting hybrid approaches could become central trends. Ultimately, 2025 may be a year that companies adapt to both the technical and logistical challenges of realizing AI’s potent.
GPU Centric Data Orchestration Becomes Top Priority
As we head into 2025, one of the challenges in AI and machine learning (ML) architectures continues to be the efficient movement of data to and between GPUs, particularly remote GPUs. GPU access is becoming a critical architectural concern as companies scale their AI/ML workloads across distributed systems. Traditional data orchestration solutions, while valuable, are increasingly inadequate for the demands of GPU-accelerated computing.
The bottleneck isn’t just about managing data flow—it’s specifically about optimizing the transport of data to GPUs, often to remote locations, to support high-performance computing (HPC) and advanced AI models. As a result, the industry will see a surge in innovation around GPU-centric data orchestration solutions. These new systems will focus on minimizing latency, maximizing bandwidth, and ensuring that data can seamlessly move across local and remote GPUs.
Companies already recognize this as a key issue and are pushing for a rethinking of how they handle data pipelines in GPU-heavy architectures. Expect to see increasing investment in technologies that streamline data movement, prioritize hardware efficiency, and enable scalable AI models that can thrive in distributed and GPU-driven environments.
Ori Saporta, vFunction
Fast AI code today will end in system gridlock tomorrow
While AI makes writing code faster, engineering teams will be challenged in 2025 and beyond to take control of their software architecture as thousands of AI-generated components interact. Teams rushing AI development will spend more time untangling messy code than writing new features. Software fixes that once took days will stretch into weeks as developers wade through AI-generated functions with hidden dependencies. Bad architecture carries many costs: skyrocketing cloud bills, increased carbon emissions, engineering teams burnout, and more.
Traditional monitoring approaches will prove inadequate as design patterns silently break down, system boundaries blur, and unexpected performance issues surface. Forward-thinking engineering teams will shift focus from code generation to deep architectural understanding, implementing new tools that monitor how AI-generated code impacts how systems evolve and detecting application design problems before they cascade. New capabilities and methodologies will be required to deal with the mass of generated code which will come with its share of AI hallucinations. Success with GenAI isn’t about writing more code faster, but about maintaining architectural integrity across application ecosystems.
Organizations must invest in next-generation observability capabilities that track architectural drift, identify service dependencies, and protect system boundaries, or risk their AI-accelerated development leading to complex, tangled systems. The winners in 2025 won’t be the fastest coders — they’ll be teams who found ways to keep AI’s speed while preventing it from turning their systems into puzzles.
Software complexity will become the bottom line: Enterprises must fix bad architecture or pay the price
Far too many organizations run bloated, complex Frankenstein systems they barely understand and can no longer sustain. The mounting pressure to increase reliability and prevent costly outages will drive companies to gain a deeper understanding of their applications and put a critical focus on optimizing their software architecture. Bad architecture carries many costs: skyrocketing cloud bills, increased carbon emissions, engineering team burnout, and more.
In the next year, to optimize applications, teams will need to have complete visibility of their software architecture to evaluate necessary services, eliminate redundancies, reduce cost and cognitive load on teams, and build applications for longevity. 2025 will be about architecting for sustainability as AI changes the course of software.
Haseeb Budhani, Rafay Systems
GenAI Will Transform Data Graveyards Into AI Goldmines
- Organizations are sitting on “data graveyards” — repositories of historical information that became too resource-intensive to maintain or analyze.
- This is largely because it can be expensive to tag data and keep track of it. Many companies defaulted to “store everything, analyze little” approaches due to the complexity and high costs related to data management.
- Yet valuable insights remain buried in emails, documents, customer interactions and operational data from years past.
- With GenAI tooling, there’s an opportunity to efficiently process and analyze unstructured data at unprecedented scale.
- Organizations can uncover historical trends, customer behaviors and business patterns that were too complex to analyze before.
- Previously unusable unstructured data will become a valuable asset for training domain-specific AI models.
Thanks to AI, Hybrid Cloud is Here to Stay
- Only about two years ago, it was a very “cloud only” environment with some companies ready to get rid of their data centers altogether.
- The reality is, many businesses still have over half their data living outside of the cloud — and it will likely stay there based on what makes the most sense for their use case (in high stakes environments such as healthcare, for example).
- Therefore, hybrid cloud strategies are alive and well, especially with the proliferation of AI.
- Organizations can maintain on-premises GPU infrastructure for consistent, high-priority workloads while using cloud GPUs for burst capacity.
- This avoids complete lock-in to cloud providers’ premium GPU pricing and grants better control over total cost of ownership for expensive AI infrastructure.
JJ McGuigan, Infragistics
Security-related attacks on AI agents will soon emerge as a critical threat: We will need more “Guardian Agents” in AI oversight
“Technology leaders will need “Guardian Agents” to autonomously monitor, manage, and contain AI actions, as they work to establish standards for AI oversight. With enterprise interest in AI agents intensifying, next-generation GenAI agents are rapidly reshaping strategic planning for product leaders. Guardian Agents will bring a holistic approach to AI security, integrating compliance assurance, ethics, data filtering, log analysis, and advanced observability. As we move through 2025, the number of product releases deploying multiple agents will rise, supporting increasingly sophisticated use cases. Guardrails, security filters, and human oversight alone won’t be enough to guarantee the safe and appropriate use of autonomous agents.
Frankie Williams, DeepL
AI will be a collaborative legal team member
AI will no longer be viewed merely as a tool but as an important team member within the legal profession, transforming the way we work and helping us work more efficiently, collaborate better, and innovate like never before. It won’t replace lawyers, but rather give us the capacity to do more of the interesting work. For instance, my team has successfully rolled out a chatbot to handle routine and repetitive customer contract inquiries, as well as general legal questions—particularly in areas where there is a lot of legal and guidance material, like privacy. This has given our lawyers more freedom to focus on more complex and strategic tasks. Being able to extract large quantities of data by simply uploading documents is also a game changer. With the help of AI, there will be many more exciting and empowering tasks for junior lawyers than the due diligence review of reams of contracts I did in the early 2000s!
Steve Rotter, DeepL
AI will accelerate hyper-personalized, more consistent marketing
Stefan Meskan, DeepL
Training and data synthesis will help break through the scaling problem
We need new ideas to move forward on the path of AI scaling laws. I see three main ways to do this: One is to improve model architectures, although I don’t expect major breakthroughs here. This has been tried a lot, and while I expect more progress in the new year, there is still a lot of steam in transformer-like architectures. Another solution is to improve optimization. Clearly, there is a lot of room to make AI training more energy and data efficient. The current approach is still very basic and consumes a lot of energy. An interesting analogy is the human brain, which consumes about 20 watts of power.
By the age of 20, this adds up to a total energy consumption of 3.5 MWh (3.5 megawatt hours). This is over 17,000 times less power consumption than training some of the most popular AI models out there! Better optimization algorithms can unlock huge efficiency gains, which is an under-explored area of research. This area is critical and will continue to be through 2025, although breakthroughs may come later.In the short run, creating more data seems like the most promising approach to further push AI scaling laws. While naive approaches to use synthetic data can hurt AI quality, with careful execution, cleverly leveraging this wealth of feedback can boost AI model performance in a wide range of tasks.
In 2025, users will shape and collaborate more with AI
There’s a lot of focus on the future of model size and technical advancements, but the real story of 2025 will come from unlocking the full potential of existing AI capabilities and enhancing human-AI collaboration. Right now, interacting with AIs is a relatively static process: you input data and receive a response.
By 2025, this interaction will become far more dynamic. AIs will not only understand users better, but will proactively offer suggestions, collaborate meaningfully, and adapt to individual needs. Many of these advanced, personalized capabilities already exist but are limited to researchers or developers. Bridging this gap and improving the user experience will be one of the most impactful advancements of the year, allowing users and organizations to create and customize their own models and interactivity. Working with an AI will increasingly feel like working with a smart coworker.
Jarek Kutylowski, DeepL
The future of AI models = tailored, custom solutions
Over the past year or two, we’ve seen much of the excitement around general-purpose AI models outpace their value – but the reality of their impact has been much more gradual. In 2025, specialized, tailored AI solutions will continue to dominate, solving specific industry challenges and delivering tangible ROI for businesses. These models are currently much more mature than general-purpose models; they’ve also been around longer, allowing more time to refine their capabilities and better align them with real-world needs.
At the same time, we also expect to see specialized models become more robust and include general-purpose aspects as part of their architecture. Looking even further into the future, I think the lines between general and specialized will blur, making room for the rise of more hybrid models with specialized and domain-specific customizations layered on top.
Sebastian Enderlein, DeepL
Voice translations will advance through contextual understanding
The next big thing for voice AI translations will be getting an even better handle on context. Right now, current systems are all about accurately perceiving spoken words. But the real challenge—and opportunity—is reasoning. Humans are great at understanding what’s unsaid through subtle cues like tone and volume… and this is where voice AI will make big leaps next year, and in the years ahead. By expanding its ability to interpret and reason about context, voice technology will be able to deliver even more seamless, intuitive interactions.
Daniel Lereya, monday.com
Productization Fuels AI Business Transformation in 2025
AI has moved beyond the hype and is now a fundamental force transforming business operations. As we move into 2025, the primary challenge won’t be about the technology but about adopting and integrating AI into existing workflows. Companies must focus on how AI can be embedded directly into platforms, including new and existing processes while extracting real and material business value to enhance and scale operations.
For AI to truly drive value, it must be accessible, predictable, and trustworthy—solutions that provide clear ROI while seamlessly aligning with how companies already work. Businesses will prioritize AI tools that grow with them and can tackle a wide range of issues, from automating routine processes to solving complex problems across areas like customer service, supply chain optimization, and data analysis—all with minimal disruption and cost.
Ultimately, success in 2025 will hinge on adopting AI and ensuring its implementation is smooth, scalable, and impactful within existing infrastructures. This will unlock new business opportunities, accelerate growth, and encourage companies to build a unique competitive edge in an increasingly AI-driven world.
Ted Krantz, interos.ai
Predictions on AI for supply chain security in 2025
Cybersecurity threats, alongside geopolitical tensions, natural disasters, global pandemic and endless other factors have made managing supply chains increasingly difficult. As the world’s supply chains continue to evolve at a rapid rate, organizations will increasingly rely on AI to ensure the security posture of their supply chain. The average organization in the S&P 500 has 1,700 direct suppliers and 1.5 million supply chain relationships through its third tier of suppliers, an 882-fold increase in relationships beyond the first tier.
AI’s ability to provide real-time risk monitoring and actionable insights will empower businesses to stay ahead of disruptions in 2025. From assessing geolocation-specific cyber risks to real-time event monitoring of cyberattacks, integrating AI into supply chain security strategies will enable leaders to shift from the reactive management to proactive threat prevention, solidifying AI’s role as a cornerstone of business continuity in the upcoming year.
Attila Török, goto.com
GenAI will be an asset, not an adversary, for CISOs
AI tools have been a double-edged sword from a security standpoint ever since their first public availability, but the focus for CISOs in 2025 should be viewing AI as an asset rather than an adversary. As these tools continue to evolve, they should be integrated into security operations to improve threat detection, response times, and predictive analytics on an ongoing basis. In a slow market, this is a material, pragmatic way to demonstrate ROI while keeping pace with the evolving threat landscape.
Stefan Meskan, Dufrain
AI will transform real-world IT management
In 2025, artificial intelligence (AI) and machine learning (ML)-based capabilities will further transform the IT support function, and lead to tangible benefits for real-world IT management at production-scale. In other words, we have all been hearing about the promise of AI, but 2025 will mark AI technologies graduating from the lab and POC environment to solve real-world problems. As IT professionals increasingly leverage AI-driven, automation to handle routine tasks, tools and technology platforms will become smarter and more advanced, emulating human expertise in basic to intermediate support and management functions.
For instance, Level 1 IT support and helpdesk roles will be increasingly augmented by AI agents and capabilities, while human workers can focus their time on more complex and value-added activities. Additionally, we can expect to see manual runbook execution and knowledge search replaced by automation and autonomous responses. This will allow for increased automated workflow capabilities with response and remediation actions generated by LLMs.
Ray Canzanese, Netskope
The Great AI Crackdown
In 2025, more leaders will realize that not everything benefits from generative AI, and so we will see a tightening of organization controls around genAI use. Organizations will consolidate their use around a few key applications that have proven benefits to the organization, for specific use cases. Applications outside of those identified will be heavily restricted, and even those core applications will have restrictions around how they will be used. This will be made even more challenging because of how much investment money is flowing into AI, which will result in everyone building AI into their applications whether or not it has any proven benefits.
Paul Laudanski, Onapsis
AI will not be a significant threat to business critical applications in 2025
I’m over the machine learning (ML) and artificial intelligence (AI) hype—it was overblown in 2024. While there are real concerns, such as mental health and misuse cases like deepfakes, it will not impact business-critical applications. When it comes to fraudulent activities or ill-intentioned use of AI, as long as companies are able to rapidly implement patches, there isn’t an increased risk to SAP security due to AI advancements.
AI has not been a significant factor in adversaries’ operations this year, even among very focused actors who know what they’re after, like nation states. If it had been, we’d already be seeing concrete results. Even for opportunistic attackers, like script kiddies, there’s nothing out there for them to package up and detonate on someone’s environment. Take the recent CISA report on the top routinely exploited vulnerabilities, for example. In 2022, SAP and Oracle were prominent on that list, but have since decreased. Although the threats are still active, this reduction reflects progress in addressing known risks, not increased activity because of AI.
What is most concerning are the SAP installations with vulnerabilities that remain unpatched, in parallel with not prioritizing the security of business-critical applications. Attackers who are interested in these apps will continue to get into your environment in other ways, not by utilizing AI – and it’s unlikely we’ll see that change in 2025.
Mohan Varthakavi, Couchbase
Businesses will adopt hybrid AI models, combining LLMs and smaller, domain-specific models, to safeguard data while maximizing results
- Enterprises will embrace a hybrid approach to AI deployment that combines large language models with smaller, more specialized, domain-specific models to meet customers’ demands for AI solutions that are private, secure and specific to them.
- While large language models provide powerful general capabilities, they are not equipped to answer every question that pertains to a company’s specific business domain. The proliferation of specialized models, trained on domain-specific data, will help ensure that companies can maintain data privacy and security while accessing the broad knowledge and capabilities of LLMs.
- Uses of these LLMs will force a shift in technical complexity from data architectures to language model architectures. Enterprises will need to simplify their data architectures and finish their application modernization projects.
AI will drive complete application rewrites as companies move beyond bolt-on solutions
- While there is now a surge of companies adding AI capabilities to existing applications, particularly in content generation and marketing, sectors like healthcare with vast amounts of untapped data will need to move beyond simple AI enhancements. Companies will realize that merely using AI to make existing applications better is insufficient, and they’ll need to completely rewrite their applications to fully capitalize on AI’s potential.
- The long-term future is a comprehensive transformation where every application – small, medium and large – is going to be revised and rewritten using AI. This sweeping movement will mark a fundamental shift from bolt-on solutions to ground-up redesigns, as organizations recognize the benefits of building truly AI-first applications that can fully harness the technology’s capabilities.
Data architectures will be redesigned to support AI integration and ensure transparency
- As AI becomes more integrated into applications, data architectures will be fundamentally redesigned to support AI workloads. Companies will implement new data architectures that go beyond simple record storage to capture the “intelligence history” and thought processes of AI systems. They will need to simplify complex architectures, including consolidation of platforms, and eliminate data silos to create trustworthy data.
- These evolved architectures will incorporate robust security measures for both data and AI communications. They will prioritize transparency and governance, enabling organizations to track how their data was used in AI training, monitor the decision-making processes of AI systems, and maintain detailed records of AI-generated insights and their underlying reasoning.
Businesses that neglect to prioritize workforce AI readiness will encounter significant challenges
- Organizations will need to develop comprehensive plans to upskill and train the existing workforce to ensure seamless integration with AI capabilities. New creative and strategic roles should be developed to complement AI capabilities rather than replacing humans with AI systems. Aggregators will play a crucial role in helping enterprises identify and implement the right AI solutions.
- Businesses must also prepare their workforce to effectively manage government AI regulations, ensuring they stay adaptable and flexible as these regulations will likely require continued updates within organizational and AI systems.
Jeffrey Wheatman, Black Kite
The AI bubble will burst, leading bad actors to pick up the pieces
It’s the golden age of AI. Nearly every cybersecurity company claims to have it and promises it’s the solution to solving security pain points while largely falling short on those promises. Next year will be the year the AI bubble bursts. AI-enabled cybersecurity companies will struggle while attackers find new ways to leverage AI for attacks, leaving defenders lagging behind. Finding credible companies with staying power in AI to help combat the increase in threats will be key for companies to keep up in the evolving threat landscape.
J-M Erlendson, Software AG
Shadow AI is here to stay
Even as companies push towards developing proprietary AI models, shadow AI will remain pervasive. People tend to favor their own way of doing things, so it’s incumbent on business leaders to evolve in how they address unsanctioned AI use.
Blanket bans may have the unintended effect of discouraging innovation, while a failure to lay out policies will bring security and compliance risks. The focus from a governance standpoint should make sure company tools are the best available options, as well as educating workers about the inherent risks of shadow AI.
AI-powered predictive analytics will evolve, driving timely decision making for businesses
Right now, AI’s capabilities in predictive analytics are still mediocre, with machine learning falling short of delivering the deep insights businesses need. While AI today mainly identifies trends, significant advancements will begin to emerge in 2025 and beyond. Over the coming years, AI will continue to evolve to provide more accurate, preemptive decision-making support, empowering organizations to act on business practices proactively and in real time, rather than giving counsel based on older context.
Proprietary data will become an AI differentiator in 2025
Generalized AI models offered a competitive advantage for those who were the first to adopt them, but implementing the tech has become a prerequisite for competing in today’s marketplace. In other words, AI is no longer a differentiator, but the way that it’s used certainly is. Companies need to keep their ‘value wedge’ (or their differences from the wider industry in the ways they do business) central to their AI strategies.
Training models on proprietary historical data attunes them to a specific organization’s nuances, yielding hyper-focused outputs and predictive analytics that are far more likely to serve business goals than blanket advice. If data is king, context is its crown, and there’s no better way to validate AI outputs than keeping its training environment airtight and focused entirely on your company.
Moshie Weis, Check Point Software
GenAI to Drive the Future of Cloud Security Against Evolving Threats
In continuation to last year, GenAI will continue to empower both attackers and defenders. Attackers can now use AI to generate complex, targeted phishing, deepfakes, and adaptive malware. In response, cloud-native security solutions leverage GenAI to automate threat detection and response across distributed environments, enabling real-time analysis and predictive defense. By 2025, using AI within cloud-native frameworks will be essential for maintaining the agility needed to counter increasingly adaptive threats.
Andrew Harding, Menlo Security
AI-driven deep fakes will become more sophisticated and hidden, bypassing traditional security measures
As Menlo Security has outlined in the Global Cyber Gangs Report in June, hyper-realistic, AI-driven cyber fraud will increase, making it difficult for individuals to discern between legitimate and malicious sites. These deepfakes will mimic trusted brands, government agencies, or even personal acquaintances, leading to automated and targeted phishing attacks and credential theft. Such attacks will largely bypass traditional security measures and exploit vulnerabilities in systems that are not yet known or patched, leading to widespread data breaches and system disruptions if enterprises don’t adopt AI-driven defenses to counter these threats.
Todd Moore, Thales
AI tools will support, not replace, security roles
AI and ML will play an increasingly central role in cybersecurity. They will be used to enhance threat detection and response (more effective anomaly detection), improve threat hunting (proactively identify vulnerabilities), combine security posture management to behavioral analytics to help monitor and secure large datasets in real-time, spotting risks such as data exfiltration attempts or unusual data access patterns.
Cybersecurity vendors are increasingly integrating AI-assisted Copilots to enhance their services for customers. These tools are great for helping to fill talent shortage gaps, which the ISC currently estimates at 4.8 million worldwide, but aren’t a replacement for internal teams. In the year ahead, it will be less about the adoption of these tools and more about how security teams leverage AI tools’ capabilities. Those looking to remain agile will likely utilize these tools to bring their threat investigation abilities to the next level.
Chene Tradonsky, LightSolver
2025 Prediction: Don’t believe the hype around the use of optical computing for AI computations
Despite the industry hype around the use of optical computing for AI computations, we anticipate faster implementation and innovation of the technology in the HPC field for complex simulations such as climate modeling and computer-aided engineering. The iterative nature of many of these computations gives optical processors a significant advantage as they can execute single calculations at a speed unrivaled by classical computers. For optical chips and systems to deliver their speed and energy-efficiency promise in AI, new methods and models must be developed and brought to maturity first, which could be a few years away.
Avani Desai, Schellman
AI-Driven Cyber Threats on the Rise
The biggest cyber threats in 2025 will stem from increasingly sophisticated, AI-driven attacks. As AI evolves at breakneck speed, attackers are deploying machine learning models that adapt, disguise themselves, and evade traditional defenses in real-time. This creates a constant race between defensive and offensive AI technologies, making it harder to detect and combat cyber threats.
Tom Keuten, Rightpoint
Data Governance will Become the Backbone of AI-Powered EX
As AI takes center stage in improving employee experience, the spotlight will increasingly fall on the integrity of data. Trust will be the key differentiator in successful AI implementations, and technologies related to data governance, quality, and explainability will be critical. With AI automating decisions and providing insights, employees and companies must trust the outputs. Building this trust will require robust data foundations that ensure accuracy, privacy, and transparency, making data governance essential for the future of AI-driven employee experience.
Hybrid Work will Evolve with AI, Rethinking Digital and In-Person Engagements
As return-to-office (RTO) policies take shape and hybrid work models become the norm, AI will redefine how employees engage both digitally and in-person. Tools like Microsoft Copilot are revolutionizing team collaboration by shifting from individual AI assistants to AI that supports group tasks. At the same time, in-person experiences will need to offer more meaningful engagement—gathering employees with a purpose rather than out of routine. Companies must balance advanced AI tools that support digital collaboration with intentional, purposeful in-person experiences that foster deeper personal and professional connections.
Jesse Murray, Rightpoint
Companies will need to customize AI tools to enhance employee experiences
Recent AI-driven expansion of collaboration tool options and capabilities is creating user confusion, lost productivity, and lower engagement. To address this trend of limitless options, companies will have to understand employees and personalize technologies accordingly, rather than employ something generic that will not stick. This includes integrating platforms with existing tools and systems.
AI-Enhanced Workflows will Redefine Employee Productivity
The next big shift in employee experience will come from AI’s ability to enhance workflows, allowing employees to focus on higher-value tasks and take on new capabilities. While we’re already seeing AI supporting tasks like note taking or generating summaries, the long-term potential lies in AI helping employees achieve tasks that were previously out of reach: designers generating code or executives extracting insights with Python, all with AI as the enabler. Over time, AI will evolve into role-specific applications that learn about employees’ individual contexts, transforming productivity across all sectors.
AI and Data Driven Insights will Drive Hyper-Personalized Employee Experiences
As companies gain unprecedented insights into how employees work, the future of employee experience (EX) lies in hyper-personalization. Tools like Microsoft Viva Insights are already analyzing digital interactions—email, meetings, and chats—to reveal key patterns in collaboration, leadership, and productivity. By combining these insights with employee engagement data from platforms like Qualtrics, employers can create tailored roles and workflows that match employees’ preferences, whether it’s flexible hours, remote work, or group collaboration. This shift will unlock new levels of employee engagement and efficiency, driving business success through truly personalized work experiences.
Rajan Goyal, DataPelago
Data Quality Supersedes Quantity, Placing a Greater Onus on AI Customers
We’re seeing growing reports that LLM providers are struggling with model slowdown, and AI’s scaling law is increasingly being questioned. As this trend continues, it will become accepted knowledge next year that the key to developing, training and fine-tuning more effective AI models is no longer more data but better data. In particular, high-quality contextual data that aligns with a model’s intended use case will be key. Beyond just the model developers, this trend will place a greater onus on the end customers who possess most of this data to modernize their data management architectures for today’s AI requirements so they can effectively fine-tune models and fuel RAG workloads.
Francois Ajenstat, Amplitude
AI investments will shift from cost-cutting to driving real customer impact
The last two years have largely been about “doing more with less,” with companies focusing on cost reduction, simplification, and technology rationalizations. But in 2025, the focus will shift toward outcomes and re-accelerating growth. After exploring the capabilities of new technologies, especially AI, businesses are now looking to make investments that actually drive value. It’s no longer just about using AI for the sake of technology– it’s about using it to deliver what customers want, how and when they want it. At its core, AI is just software. While it can be incredibly powerful, it’s only valuable when it solves real customer problems. More organizations are recognizing this shift and focusing on the right investments that deliver tangible impact.
Casey Ciniello, Infragistics
Implementing AI Will be a Top Priority in 2025
Ariel Katz, Sisense
The Demise of Traditional BI: API-First and GenAI Integrate Analytics into Every App
In 2025, traditional BI tools will become obsolete, as API-first architectures and GenAI seamlessly embed real-time analytics into every application. Data insights will flow directly into CRMs, productivity platforms, and customer tools, empowering employees at all levels to make data-driven decisions instantly—no technical expertise needed. Companies that embrace this shift will unlock unprecedented productivity and customer experiences, leaving static dashboards and siloed systems in the dust.
The Semantic Layer Becomes the Enabler for LLMs in Enterprises
In 2025, the Semantic Layer will become the crucial enabler for LLMs in enterprises, acting as a bridge between internal data and LLMs to deliver precise, contextually relevant insights. By unifying enterprise data with global knowledge, this integration will revolutionize decision-making and productivity, making GenAI indispensable. Companies that embrace this convergence will dominate in innovation and customer experience, leaving competitors behind.
Mehdi Daoudi, Catchpoint
DevOps Supercharges AI-First Infrastructure in 2025
DevOps will evolve to meet the unique demands of AI-driven infrastructure, where complex ecosystems of data, machine learning models, and interconnected systems power nearly every industry. This AI ecosystem involves managing vast amounts of data, training and deploying machine learning models, and supporting scalable compute resources—all requiring specialized infrastructure. DevOps teams will expand their role, going beyond workflow automation to fully owning and optimizing these AI-first infrastructures. They’ll set best practices for managing the speed, scale, and reliability of AI applications, helping organizations harness AI efficiently and securely as it becomes central to operations.
Bill Bruno, Celebrus
We will continue to see great advancements for AI in industries where data is plentiful, such as healthcare, but we’ll see brands struggling to activate AI in meaningful ways for their consumers
Most of this will be driven by the discovery that most of their data is unstructured, incomplete, and is full of biases due to how digital data has been captured over time on their websites and apps. We will see a rise in stories of poor uses of AI as a result as well, which will cause brands to pump the brakes a bit and revisit their data strategies.
Philip George, InfoSec Global Federal
AI’s Role in Assessing and Creating New Quantum-Safe Algorithms
An immense amount of time, effort and checking and double-checking complex math went into creating the first three encryption algorithms published by NIST. Those involved in the project are eager to improve and expedite the process going forward. In an age when everyone is considering how AI can play a role in improving various processes, of course the question has come up for quantum-safe algorithms as well.
However, there is a long way to go before this will be put into practice at any level. Checks and balances for AI must be determined before we can assume that AI has adequately tested a new algorithm, or has created an algorithm that is mathematically sound not just in theory but in practice as well. More discussion around this topic will take place in 2025, and likely some experimenting, but undoubtedly, at some point in the future, we will see AI take a more active role in developing and accessing new sets of quantum-safe algorithms.
Gilad Shriki, Descope
AI will continue to be leveraged as a prominent attack vector for cybercrime
We’ll see a surge in fraud schemes where threat actors use AI to impersonate legitimate parties. At the same time, attacks against user-facing AI will rise because of their inherent vulnerability. Cybercriminals will attempt to “jailbreak” or social engineer their way past security protocols, which will drive the need to protect or limit AI agents from unauthorized access and manipulation.
Rishi Bhargava, Descope
AI will get a major authorization upgrade (or will require one)
Today’s simple permission models won’t scale for AI systems that can generate code, access sensitive data, and interact with users in increasingly sophisticated ways. In 2025, organizations will need to build context-aware authorization that protects against the unique vulnerabilities of AI systems.
Additionally, user experience will make or break AI apps: When every application has AI capabilities (and they soon will), the differentiator will be seamless user experience. Companies that force traditional authentication checkpoints into AI interactions will see significant dropoff. The winners will be those who make security feel like a natural part of the conversation.
Lastly, passkeys will reach critical mass: With major platforms completing their passkey rollouts in 2024, 2025 will be the year that passkeys become mainstream for everyone else. As SMEs gain access to better implementation tools and users grow more comfortable with biometric authentication, passwords will finally begin their long-overdue retirement.
Stephen Manley, Druva
We’ll see the first data breach of an AI model, temporarily refocusing efforts in favor of shoring up security vulnerabilities.
Paige Schaffer, Global Identity & Cyber Protection
2025 Cybersecurity Predictions – A Rise in Deepfakes
Advancements in AI have already allowed criminals to create highly convincing deepfake content, opening the door for new forms of deception and fraud. In particular, deepfakes could be used by scammers to trick victims into handing over money by impersonating a trusted friend or family member. On the business side, deepfake technology can also be used in elaborate social engineering schemes.
Itamar Golan, Prompt Security
The Future of Work with AI
Contrary to widespread concerns, I don’t expect AI to eliminate jobs in 2025. Instead, it will serve as a powerful tool to enhance human capabilities. Agentic AI systems will work alongside humans, like in customer service, sales outreach, marketing content creation, software development and healthcare applications, among others. This means that very soon, 30% of our tedious and repetitive tasks will be automated, giving us more time to focus on creative, innovative and interesting pursuits.
I believe we will also see a significant shift as the multi-modality of AI becomes more mainstream (video, audio, etc.), as opposed to the majority of the use of AI which has been text-based. This creates new opportunities for human-AI collaboration.
Organizational AI Adoption
The democratization of LLM access, driven by ever-decreasing prices, is enabling broader adoption across organizations. Additionally, specialized AI solutions will increasingly be moving away from OpenAI’s dominance, with alternatives like Claude gaining traction in specific domains such as coding, which is something we’re already starting to see.
Agentic AI
AI chatbots use generative AI to provide responses based on a single interaction. A person makes a query, and the chatbot uses natural language processing to reply.
In my opinion, the next frontier of artificial intelligence will be agentic AI, which employs sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems. It is poised to enhance productivity and operations across various industries.
Agentic AI systems process vast amounts of data from multiple sources to independently analyze challenges, develop strategies, and execute tasks such as supply chain optimization, cybersecurity vulnerability analysis, and assisting doctors with time-consuming tasks.
I believe that by 2025, we will see a significant increase in resources shifting from single-interaction procedures with LLMs to this multi-step approach of agentic AI, which will gradually solve complex problems for us autonomously.
Itamar Golan, Prompt Security
The Future of Work with AI
Contrary t
Tim King
Executive Editor
Tim is Solutions Review's Executive Editor and leads coverage on data management and analytics. A 2017 and 2018 Most Influential Business Journalist and 2021 "Who's Who" in Data Management, Tim is a recognized industry thought leader and changemaker. Story? Reach him via email at tking@solutionsreview dot com.
- Storage and Data Protection News for the Week of January 10; Updates from Broadcom, Cohesity, KnowBe4 & More - January 10, 2025
- Data Management News for the Week of January 10; Updates from Anomalo, Dremio, Oracle & More - January 10, 2025
- Analytics and Data Science News for the Week of January 10; Updates from Domo, MicroStrategy, Reddit & More - January 10, 2025
