March 4, 2025. Artificial intelligence (AI) is evolving at an astonishing pace from a new technology to a business influencer. From developing AI agents to interacting with technology that feels more like a natural conversation, AI technologies are poised to transform the way we work.
Categories
Tags
Contact info
Silicon Saxony
Marketing, Kommunikation und Ă–ffentlichkeitsarbeit
Manfred-von-Ardenne-Ring 20 F
Fax: +49 351 8925 889
Contact person:
But what exactly is in store for us? We would like to introduce you to five topics that will be very important for AI in 2025. These will certainly pose challenges for companies, but they also have the potential to push the boundaries of what is possible. Are you ready for a glimpse of what awaits us in the new year and beyond? Let’s go!
AI agents are still in their infancy. Many software providers are just launching their first so-called “AI agents”, which are only based on a simple dialog-oriented document search. However, the trend is already moving towards innovative AI agents that are able to plan, draw logical conclusions, use tools, collaborate with humans and other agents and iteratively reflect on progress until they have reached their goal. In 2025, these agents will develop rapidly and act increasingly autonomously. More specifically, in 2025, AI agents will be used more in the background to manage complex agent-assisted workflows.
Users will interact with an assistant to support their tasks, which will implement the request and coordinate between systems of multiple specialized AI agents to complete more difficult tasks. Future AI agents or multi-agent systems (MAS) will be able to work together to understand users, have full context and structure a problem. They then interact with these domain-specific, specialized AI agents, each performing specific subtasks that together contribute to completing a much more complex task. In the future, users will no longer even have to trigger an action. Instead, AI agents will proactively respond to business events such as incoming customer inquiries, supply chain disruptions or demand peaks. They will automatically prepare a decision workflow as much as possible before involving human users for feedback.
In the next five years, AI agents will simplify significant parts of workflows, including aspects that are currently difficult to automate. These include exceptions in customer service, tedious administrative tasks and special programming activities such as coding or debugging software. AI agents will be flexible. They will be able to plan, fail and try something else or correct themselves based on conclusions. AI agents will perform repetitive routine tasks as effectively and often more effectively than humans. This will lead to higher productivity and demonstrable cost savings. For tedious and very extensive tasks, agents will be more adaptable and robust than traditional robotic process automation (RPA). This means that they will be able to find the best out of many possible outcomes, which is almost impossible to program in an RPA algorithm using traditional automation methods.
The introduction of AI in these areas will also change the dynamics within the workforce. The role of humans will increasingly focus on anticipating non-routine scenarios, dealing with ambiguity, accounting for human behavior, making strategic decisions and driving real innovation – augmented, not replaced, by AI capabilities.
In short, AI will take over extensive routine tasks, while the value of human judgment, creativity and high-quality results will increase.
The trend of large language models (LLMs) increasingly serving as raw material for routine generative AI tasks will continue. LLMs are increasingly drawing on a pool of public data from the internet. This will intensify. And companies will have to learn to adapt their models to unique data sources with rich content. Model improvements in the future will not be achieved by brute force and larger amounts of data, but by better quality data, more context and the refinement of underlying techniques. Companies need to invest more time in innovation to develop better models through fine-tuning and model adaptation, rather than training ever larger models. Neurosymbolic AI techniques, especially knowledge graphs, will experience a renaissance as they can provide both learning goals for base models and context to significantly improve the performance of generative AI while reducing hallucinations.
We will also see a greater variety of base models for different purposes. Take, for example, Physics-Informed Neural Networks (PINNs). They generate results based on predictions based on physical reality or robotics. PINNs will become increasingly important in the labor market as they enable autonomous robots to navigate and perform tasks in the real world. Possible applications range from warehouses to production plants. They are also suitable for models that are trained with tabular, structured data such as the SAP Foundation Model and can handle tasks that LLMs are not as well suited for, such as predicting numerical values.
Models will become increasingly multimodal, meaning that an AI system can process information from different types of input. AI applications will eventually evolve into “any-to-any” modality solutions that are able to understand, process and draw conclusions from text, voice, image, video and sensor data using a single model. In addition, smaller and more specialized LLMs with scalable techniques for fine-tuning and the ability to work on any device will become more common. This trend may lead to highly personalized models for businesses or even individuals in the future.
Businesses will turn to strategies that leverage multiple base models (not to be confused with the multimodal capabilities of a single model described earlier), using a range of AI models and techniques tailored to specific use cases. This is supported by the trend towards optimizing small parts of models, which requires fewer resources and much less data. This creates fully flexible models that allow companies to extract more value from their unique data and gain a competitive advantage. Enterprise software providers will offer or extend marketplaces and platforms with integrated AI models that support seamless model deployment, management and updating. Benchmarking and the reduction of model change costs will make it easier to deploy the same use cases in heterogeneous environments. Context equals benefit. Knowledge graph technology has been around for 40 years. Now it is experiencing a revival as it can be used to overcome key challenges of LLMs, such as understanding complex formats, hierarchies and relationships between business data. Knowledge graphs give meaning to data and explain the relationships between entities. This significantly improves the capabilities of LLMs. The next step in this journey is large graph models, which will enable new advances in generative AI.
Implicit knowledge is power – and making knowledge explicit to others is an even greater power.
While 2024 was all about the introduction of AI use cases and their benefits for companies and also individuals, 2025 will be the year of unprecedented adoption of enterprise-specific AI. More people will understand when and how to use AI, and the technology will mature to the point where it can tackle critical business problems such as complex multinational affairs. Many companies will also gain hands-on experience as they face AI-specific legal and privacy regulations for the first time (compared to 10 years ago when companies started moving to the cloud), laying the groundwork for them to apply the technology to their business processes.
In technological terms, significant progress was made in AI in 2024. In 2025, however, companies will focus on making more sense of these advances through seamless data integration to ultimately improve the accuracy and meaningfulness of AI-powered results and increase adoption. Last but not least, it is conceivable that in 2025 we will see a shift in focus in the business model with software: from the creation of static software functions to an outcome-as-a-service model focused on achieving process goals.
The next frontier for AI is to seamlessly bring people, data and processes together to improve business outcomes. In 2025, we will see increasing adoption of AI at all staffing levels as people discover the benefits that can come from human performance plus AI.
This means that the classic user experience will shift from system-driven interactions to intent-based communication driven by human dialog, with AI active in the background. The new user interface for interacting with a system will consist of AI assistants. This will make software more accessible to humans. AI will not be limited to a specific application, it could even replace it one day. AI will blur the boundaries between frontend, backend, browser and applications. It’s a bit like equipping your AI with “arms, legs and eyes”. Power users will still work with individual, very specific interfaces, but most users will demand flexibility across multiple access patterns. At the same time, there will be an increased willingness to accept longer inference times in order to obtain high-quality answers to complex, previously unsolvable problems and actions in areas that require in-depth analysis and research. Users will ultimately recognize the trade-off between the latency and complexity of tasks handled by AI.
Importantly, companies will no longer view AI as just a collection of productivity tools, but will realign their workforce as a network of collaboration-based intelligence where AI agents and humans work together to accelerate innovation in the enterprise. The combination of human expertise in strategic thinking with the strengths of artificial intelligence in comprehensive analysis and pattern recognition will give companies new competitive advantages – if they can effectively coordinate such networks of hybrid intelligence and thereby promote groundbreaking discoveries and market opportunities. Next year will also mark the early stages of another significant change in the way humans and AI work together: AI agents will evolve into workflow partners, taking the first steps toward autonomously navigating software environments and automating routine tasks – from data analysis and report generation to scheduling coordination and software testing. This will also be the start of a longer transition towards changing work processes and patterns, with forward-thinking companies developing new roles, metrics and training approaches for effective collaboration between humans and AI.
It could be argued that governments around the world are struggling to keep pace with the rapid advances in AI technology and create meaningful regulatory frameworks that provide appropriate guidance for AI without restricting innovation. The regulatory landscape will become even more heterogeneous; the OECD AI Policy Observatory is currently tracking hundreds of AI-related regulations being discussed around the world. This requires an assessment of compliance with different regulatory frameworks for models and the technical interpretation of these frameworks.
By 2025, the discussion will shift from what we are trying to regulate from a technical perspective to how we create innovation and what we consider to be fundamentally human. This discussion will elevate the role of humans, to a much more positive perspective and also help us develop a long-term vision for how we envision humans and AI living and working together.
In this environment, it will continue to be crucial for companies developing and using AI technology to adhere to responsible principles in terms of safety and ethical use. This will also help to set the framework for important precedents and compliance.
However, these are just a few examples of the many exciting advances we believe AI will make in 2025. Overall, the key takeaway from the coming year will be to increase the usability of existing breakthrough technology. We will see AI becoming much more deeply and almost invisibly embedded in consumer and enterprise applications, and how vendors and companies using these applications will seamlessly integrate their individual frameworks and data with AI.
In terms of the overall use of AI, however, companies will need to adopt a modern cloud suite with unified data access and harmonized data models. Only then can they overcome data silos and fully benefit from AI innovations that span the entire organization. This will significantly improve the accuracy and meaningfulness of AI-powered results and ultimately increase adoption, especially within organizations.
Sean Kask is Vice President and Head of AI Strategy for SAP Business AI at SAP.
Walter Sun is Senior Vice President and Global Head of AI for SAP Business AI at SAP.
Jonathan von Rueden is Head of AI Frontrunner Innovation for SAP Business AI at SAP.
– – – – – –
Photo: pixabay
