As AI advancements continue unabated, the search for ROI (return on investment) will intensify, according to experts. They predict that 2025 will be the year of agentic AI, but it will also be the year of AI pragmatism. These twin expectations mark a maturing of the AI landscape and dampening of the hype from two years ago, when ChatGPT first came on the scene.
Here are the top AI trends for 2025, according to IBM, Google Cloud, SAS, Cloudflare and Wix.
1. Agentic AI is here – and building guardrails is a must.
As agentic AI emerges as a predominant theme in 2025 – marking a fundamental shift from traditional AI tools to proactive agents and teams of agents – so too will questions around accountability and control of these increasingly autonomous systems. This will bring greater attention to the guardrails, processes and tools for how we govern agents, in order to build trust for this powerful new frontier of AI capabilities. It will also heighten the need to upskill employees across every discipline and leadership level so they can responsibly develop, use and oversee agentic solutions
—Ritika Gunnar, general manager, Data & AI, IBM
2. Enhanced human-AI collaboration. As AI grows more powerful, daily workflows will increasingly incorporate agentic AI activities with human oversight. For instance, in customer support, AI may handle routine inquiries while humans tackle complex issues; in software development, coders design the architecture while AI implements much of the underlying logic. This approach ensures a transparent, trust-driven partnership that combines human creativity and direction with AI’s speed and scalability.
—Eli Brosh, head of AI Research at Wix
3. The AI revolution will hinge on edge computing
To unlock AI’s true potential, edge computing must bring the compute power closer to where it’s actually needed. Edge computing represents a paradigm shift, dramatically reducing latency and enabling a new generation of sophisticated, responsive applications. Imagine autonomous vehicles making split-second decisions, interactive gaming with zero perceptible delay, and real-time video processing that responds instantaneously. These innovations become possible when compute resources are strategically positioned near their point of use. That’s why the future of AI is not just about raw computational power, but about smart, distributed computing that brings intelligence closer to where it’s most impactful.
—John Engates, field CTO, Cloudflare
4. Open-source AI will push enterprise adoption over the hump.
Many enterprises are struggling to show measurable returns on their AI investments — and the high licensing fees of proprietary models are a major factor. In 2025, open-source AI solutions will emerge as a dominant force in closing this gap. Thanks to their community-driven development, open-source models are quickly equaling the major proprietary offerings in power, and the proliferation of open industry- and task-specific AI solutions will make it easier than ever for organizations to apply them to a wide range of innovative use cases without hefty fees or API call costs.
— Bill Higgins, watsonx Platform Engineering and Open Innovation, IBM Research
5. 2025 will be the year of AI pragmatism. After a period of experimentation, organizations will now be more value-conscious with their AI spend. Organizations will be more scientific and methodical in how they approach putting AI in front of customers, evaluating different approaches and options for different use cases. We’re seeing teams pivot to predictable pricing models, transparent gateway metrics, and smaller models that do the job, rather than the largest, most expensive LLMs.
—Rita Kozlov, vice president, product management, Cloudflare
6. LLMs get commoditized … and specialized
In 2025, LLMs will become commoditized, leading to AI pricing models collapsing as base-level capabilities are offered for free. The real value will shift to specialized services and domain-specific applications built on top of these models. Simultaneously, the rise of open-source LLMs will challenge the dominance of a few key providers, driving a more decentralized AI landscape where customization and integration will be the key differentiators.
—Udo Sglavo, vice president, Applied AI & Modeling, R&D, SAS
7. Multimodal AI, especially for complex document processing, will grow significantly within the enterprise. Multimodal AI is poised to drive substantial value to enterprises, enabling them to unlock more value from their data. Multimodal AI models are capable of processing and analyzing all manner of complex documents with embedded rich content in the form of images, tables, and charts. These models are also evolving to support other modalities such as audio and images, unlocking countless new possibilities for insights. As a result, organizations will need to begin bringing order and method to the way they handle all of this multimodal unstructured data to get it ready for enterprise AI. This will put pressure on existing infrastructure, including greater storage requirements and robust management solutions.
—Sriram Raghavan, vice president at IBM Research for AI
8. Financial institutions are increasingly using AI to combat the rising threat of fraud, including AI-generated attacks, which are on the rise. By analyzing unstructured data, identifying complex patterns, and automating threat detection, AI will become the ultimate shield against financial fraud, and crucial in detecting and preventing cyberattacks, protecting sensitive customer data, and maintaining trust.
—Zac Maufe, Google Cloud’s global head of regulated industries
9. AI will battle AI in this new era of cybersecurity. AI powers advanced threat detection, anomaly detection, and automated response systems, enabling defenders to stay ahead of emerging threats. On the other hand, it is also being weaponized by attackers to create more sophisticated and adaptive exploits. We are entering an era where AI systems will battle AI systems, with human security teams orchestrating strategies to maintain the upper hand. This shift underscores the need for continuous innovation in AI-driven security solutions, as static defenses become increasingly inadequate.
—John Engates, Field CTO, Cloudflare
10. The rise of shadow AI
Enterprises have long dealt with Shadow IT – the use of non-approved Cloud infrastructure and SaaS applications without the consent of IT teams, which opens the door to potential data breaches or noncompliance. Now, enterprises are facing a new challenge on the horizon: Shadow AI. Shadow AI has the potential to be an even bigger risk than Shadow IT because it not only impacts security, but also safety. The democratization of AI technology with ChatGPT and OpenAI has widened the scope of employees that have the potential to put sensitive information into a public AI tool.
—Nataraj Nagaratnam, CTO, IBM Cloud Security
Be First to Comment