To navigate AI’s evolution, organizations need to understand that it’s no longer just a question of “What can AI do?” but “How do we set our organization up for success with AI? How do we build for it? What problems do I solve with which models? How do we govern it?”

Looking ahead to five critical themes that will define enterprise AI in 2026, these present both opportunities and challenges for organizations. Let’s dive in.

1. New categories of AI foundation models unlock enterprise value

Advances in generative AI stem from breakthroughs in “foundation models,” massive neural networks trained on vast amounts of data that can be adapted to a wide range of tasks.

Large language models (LLMs) were the first wave of foundation models at scale. General-purpose LLMs, trained on the equivalent of all the text on the internet, opened the door to many value-adding use cases, including summarizing documents, writing code, and powering applications like ChatGPT and Claude. Over the last few years, we have already seen the foundation model approach applied to other domains, such as video creation and voice.

In 2026, specialized foundation models optimized for specific data types and domains will power the high-value enterprise AI use cases. Video generation models have already shown that models grounded in real-world physics data can reason about scenes and physical dynamics. Emerging world models demonstrate that simulating the physical world unlocks new possibilities in simulation, synthetic training data, and digital twins. Vision-language-action models demonstrate that robot-specific foundation models can generalize to new tasks and environments, enabling the transformation of web-scale knowledge into real-world actions in logistics and manufacturing.

In the enterprise domain, a similar shift is underway for structured data found in databases and transactional business software. While LLMs are impressive across many enterprise use cases, they cannot handle tasks like numerical predictions, such as inferring a delivery date or supplier risk score. However, work on relational foundation models shows that training on structured datasets—for example, data in tables, rather than generic text or images from the internet—can deliver high predictive accuracy without the tedious feature engineering and training required in classical machine learning. This means organizations can deploy predictive models in days, not months. Recent launches of relational foundation models, such as SAP-RPT-1, Kumo, and DistilLabs, highlight how new models can directly support use cases like forecasting, anomaly detection, and optimization across ERP, finance, manufacturing, and supply chain scenarios.

In 2026, these specialized models are expected to scale to deliver superior performance and economics for structured business tasks, surpassing general-purpose LLMs and state-of-the-art machine learning algorithms. These models will emerge as the workhorses behind high-value enterprise tasks.

2. Software evolves toward AI-native architecture

AI has seen various approaches create value over the decades, from the first rules-based expert systems to probabilistic deep learning and the recent explosion in generative AI. In 2026, organizations will shift from enhancing existing AI applications and processes to AI-native architectures, which will fully realize the promise of modern AI.

AI-native architecture adds a continuously learning, agentic intelligence layer on top of deterministic systems, enabling applications to become intent-driven, context-aware, and self-improving rather than being statically coded around fixed workflows. Agentic systems will still only be as good as the context layer they can reliably retrieve and ground on. Here, organizations should invest in truly comprehensive, semantically rich knowledge graphs that provide a scalable source of context, making AI-native software dependable and self-improving.

Enterprise applications will increasingly be built natively around AI capabilities, featuring user experiences designed for multi-model, natural language interaction; AI agents reasoning through complex processes; and a foundation managing foundation models, services, and a knowledge graph capturing semantically rich business data.ĚýAI-native architecture also enables more employees to create apps—such as smaller, ad-hoc productivity applications—in a matter of minutes without straining IT.Ěý

AI-native architecture builds on, and even requires, established SaaS principles and investments in modern cloud applications. The technical term for combining probabilistic, adaptive AI models with deterministic systems of record is called neurosymbolic AI. It brings together AI’s best capabilities to adapt with reliable, governable, and deterministic processes. Next-gen applications will not just have AI bolted on; they’ll be built around AI at their core. This means combining reasoning, business rules, and data to deliver insights and automation seamlessly. Imagine ERP systems that proactively flag anomalies, recommend actions, and even execute workflows autonomously—all while staying aligned with company policies and regulations.

3. Agentic governance becomes mission-critical

Over the past two to three years, generative AI has introduced a wave of value-added use cases. These use cases were largely based on users sending a prompt to a model, receiving a response, and then interacting with the model again.

Last year saw the start of the next wave of innovation: AI agents capable of planning and iteratively reasoning through multi-step tasks, including selecting tools, self-reflecting on progress, and collaborating with other AI agents. These advanced AI agents promise to tackle complex business processes that were previously immune to automation, such as analyzing myriad documents, records, and policies to or .

However, the proliferation of AI agents, many of which handle critical tasks and sensitive data, demands the development of new capabilities. Agentic governance will emerge as a critical capability as organizations deploy hundreds of specialized AI agents. The “agent sprawl” challenge will mirror previous shadow IT crises, but with higher stakes given agents’ autonomous decision-making capabilities.

Forward-thinking enterprises will establish comprehensive governance frameworks addressing five dimensions: agent lifecycle management (version control, testing protocols, deployment approval, retirement procedures); observability and auditability (agent inventory, logging, reasoning paths, and action traces); policy enforcement (embedding business rules, regulatory constraints, and ethical guidelines into agent execution); human-agent collaboration models (defining autonomy boundaries, approval requirements, and escalation pathways); and performance monitoring (tracking accuracy, efficiency, cost, and business impact).

The organizational shift will prove profound—from viewing AI as an independent tool to managing agents as digital coworkers requiring onboarding, performance reviews, and continuous improvement. HR and IT functions will collaborate on “digital workforce management” as organizations treat agentic governance as seriously as they do traditional workforce oversight.

4. Intent-driven ERP and generative UI emerge as a new user experience

Consumers are becoming increasingly familiar with computer interactions requiring prompts in natural language, voice, and even images and gestures. At the same time, generative AI’s ability to create text, graphs, code, and HTML on the fly is improving rapidly. In parallel, AI agents enable users to simply express their intentions, allowing the agent to determine how to work toward achieving that goal.

These advancements open the door to varied and entirely new modalities for users to work with enterprise software, as well as “no-app ERP” experiences. For example, to book a customer visit, a worker typically needs to open an analytics application to review the account, look in the CRM system to retrieve the customer’s address, and then navigate to another application to book travel, among other tasks. 

In 2026, we will see “gen UI” experiences increasingly surface via digital assistants, relieving users from the need to navigate between multiple applications and perform manual tasks. With time, AI will allow the user to simply express the intent: “Prepare a trip to my customer with the most leads.” From here, an AI agent will plan out the steps and required systems, interacting with the user to confirm travel details while dynamically generating analytical graphs and briefing material in the window. As AI agents develop stronger calculation and prediction tools, users will be able to “speak to their data” more naturally, with agents making data-based decisions in the background. To be clear, interactions with agents will extend far beyond a chat box; organizations will enjoy rich visualizations, complete workflows, and the ability to build hyper-personalized apps with just a few commands.

The user interface will not disappear. No-app ERP experiences and autonomous agents require the same foundational substrate that humans rely on for their daily work: structured workflows, security, governance, and business logic defined in business applications. The difference is that agents consume these primitives programmatically at scale, not only through a GUI, and humans can interact with these agents via natural language without ever needing to open the application.

These capabilities will usher in a new paradigm for human-AI collaboration and productivity in the workplace. Personalized experiences and adaptive workflows across applications and data sources will lower adoption barriers. This ability to focus solely on achieving a user’s intention, regardless of the interaction modality and underlying systems, will drive return on investment (ROI) in AI and enterprise software.

5. Deglobalization drives sovereign AI offerings

AI sparked debates about digital sovereignty among nations due to AI’s potential impact on everything from scientific discovery and national security to economic productivity and even culture. Events in geopolitics, such as supply chain disruptions caused by tariffs and war, have only intensified the urgency that many nations and organizations feel to become digitally sovereign.

Digital sovereignty has two broad definitions. First, digital sovereignty is an information security designation governing data storage and access, such as U.S. FedRAMP and German VSA, required to process sensitive governmental data in a “sovereign cloud.” Second, and more broadly, sovereignty refers to the provenance of physical assets, intellectual property, legal jurisdiction, and services along the cloud stack. For example, does an application utilize an AI model created in Europe, the U.S., or China, and is the data center geographically isolated? 

The high stakes, geopolitical uncertainty, and complexity of “sovereign AI” will lead enterprises to increasingly demand AI and cloud solutions that are simultaneously cutting-edge, flexible, and fully sovereign. This intensifies the shift from globalized one-size-fits-all cloud to regionally compliant, AI-powered enterprise platforms. At the same time, governments will continue to refine their national AI strategies to invest in areas along the stack where they can compete and create value.

Executing on the 2026 AI themes

In 2026, AI is poised to move from a supporting tool to a fundamental pillar of the enterprise. This shift is driven by a convergence of defining trends—including increasingly capable agents, generative UI, and AI-native architecture—that push AI from the application layer and into the very core of business operations.

Organizations that thrive will be those that recognize this shift and build an enterprise that is purpose-built for AI: establishing robust governance to manage a new, collaborative workforce of humans and AI agents; embracing gen UI to lower adoption barriers and an intent-driven user experience that helps employees interact naturally; seeking out specialized foundation models that are precisely tuned for enterprise use cases to drive business value; and, finally, building applications natively around AI that combine reasoning, business rules, and data, delivering proactive insights and automation.

However, in 2026, organizations will still need high-quality, connected data. Data siloes severely limit the effectiveness of AI. As mentioned, AI-native architecture requires established investments in modern cloud applications that harmonize data across the entire business—because unified data means AI’s outcomes are more accurate and relevant.

Jonathan von Rueden is chief AI officer at 51·çÁ÷SE.
Walter Sun is senior vice president and global head of AI for 51·çÁ÷Business AI at SAP.
Sean Kask is vice president and head of AI Strategy for 51·çÁ÷Business AI at SAP.

But as AI becomes more independent, new risks emerge. So, how can we navigate this next frontier responsibly? This is a question that we at 51·çÁ÷do not leave to chance.

From tools to teammates

Imagine you’re buying a car. You expect it to meet all safety standards, regardless of where the component parts are built or how the car is assembled. The process behind the scenes does not change your expectation of safety. The same goes for agentic AI.

Agentic AI systems are more than tools; they are intelligent agents that plan, learn from experience, self-correct, and collaborate. They’re capable of orchestrating complex processes, making decisions, and even engaging with other agents or humans to achieve a goal. However, with this leap forward comes a new layer of complexity and risk.

Core capabilities and risks of agentic AI

Agentic AI systems bring powerful capabilities like planning, reflection, and collaboration, enabling them to tackle complex tasks autonomously. They can map strategies, learn from mistakes, use external tools, and coordinate with humans and other agents.

However, each strength introduces risks. For example, flawed planning can cause inefficiencies, reflection may reinforce unethical behavior, tool usage can lead to instability when systems interact unpredictably, and unclear collaboration can result in miscommunication and compounded errors. Balancing these capabilities with proper safeguards is essential for safe, ethical deployment.

Managing autonomy: balancing freedom with control

One of the most pressing challenges with agentic AI is managing its autonomy. Left unchecked, these systems can veer off course, misinterpret context, or introduce subtle risks without immediate detection. To address this, organizations must strike a careful balance between freedom and control.

We have learned that oversight should be calibrated according to risk. High-stakes domains like healthcare or human resources demand robust human supervision, while low-risk, routine tasks can tolerate greater autonomy. Also, continuous monitoring is essential; agentic AI systems, like any complex technology, require regular checks to ensure quality, compliance, and reliability.

A key element of this oversight is maintaining a “human in the loop” approach, where human judgment is integrated into critical decision points, ensuring that automated actions remain aligned with human values and organizational intent.

This principle has been at the heart of SAP’s ethical AI approach from the beginning, reflecting our belief that AI should augment, not replace, human decision-making. To reinforce this, 51·çÁ÷has introduced mandatory ethics reviews for all agentic AI use cases, ensuring that each deployment is scrutinized for ethical implications and remains aligned with our responsible AI principles.

Click the button below to load the content from YouTube.

Always allow YouTube

Building transparency and accountability

Transparency is not just a buzzword; it’s a foundational requirement for building trust in agentic AI. From the outset, during the design phase, it is crucial to classify AI systems based on the complexity and risk of the tasks they perform. This classification guides decisions about the necessary safeguards and ensures that mechanisms for human intervention are integrated from the beginning.

At runtime, transparency is maintained through explainability and traceability. Developers and end-users must be able to understand what the system is doing and why. Crucially, accountability must always rest with humans or legal entities, never with the AI itself.

Rethinking governance and regulation

Despite the emergence of agentic AI, there have been no new regulations specifically crafted for it. Existing laws and frameworks such as GDPR still apply and provide a solid foundation for governance. However, what has changed is the level of technical rigor required to remain compliant and ethically sound. Organizations must now adopt more robust processes. They need to analyze use cases with greater precision, apply risk-based controls that match the potential impact of the AI system, and ensure that ethical and legal standards are upheld through enhanced design practices and ongoing testing.

Designing with human values at the center

Agentic AI cannot be an excuse for lowered standards. At SAP, the stance is unequivocal: Even in autonomous systems, AI must meet the highest ethical benchmarks. This means embedding principles such as fairness, transparency, and human agency directly into the design.

Ultimately, all users should be equipped with the tools and understanding they need to supervise and, when necessary, intervene in the system’s behavior.

Building trust in a black-box world

Trust in AI doesn’t happen by default; it must be intentionally built and continually reinforced. One of the most effective ways to do this is by giving stakeholders the right amount of information. Too much detail can be overwhelming and counterproductive while too little fosters blind trust or fear of the unknown. The key lies in communicating clearly about the system’s capabilities, risks, limitations, and appropriate use. Empowering users to critically assess the AI’s behavior – and to know when to step in – is central to creating a safe, secure, and trusted AI environment.

Rethinking KPIs in the AI-augmented workplace

As agentic systems, like our Joule Agents, begin handling more tasks, human roles will naturally evolve. To keep up with this shift, organizations need to rethink how they define and measure success. This starts with investing in change management and upskilling programs that prepare employees to work effectively alongside AI. It also requires redefining productivity metrics, moving beyond task completion to focus on how well humans and AI agents collaborate. Success should be measured by how efficiently teams harness AI to unlock new levels of insight and innovation.

Building AI that builds trust

Agentic AI is not just another phase; it is a transformation. But like any transformative technology, success depends on how it’s built, governed, and used.

At its best, agentic AI amplifies human capabilities, accelerates innovation, and helps tackle challenges once considered too complex. But it also demands a new level of diligence, oversight, and ethical reflection.

The future is not just about building smarter agents; it’s about building responsible ones.

Learn more:

• The Evolution from Chatbots to Agentic AI Systems

Walter Sun is senior vice president and head of AI at SAP.

That is why 51·çÁ÷is excited to announce our expanded partnership with Cohere, a leader in secure, enterprise-grade AI.

Together, we plan to bring Cohere’s powerful generative and advanced retrieval models to the 51·çÁ÷ecosystem, starting with its model, and extending evaluations with , to enrich our product suite, playing an important role in powering agentic AI experiences.

These models are planned to be available alongside other leading AI models from 51·çÁ÷as well as third parties in the generative AI hub in 51·çÁ÷AI Core, with the intent to give customers more choice to build AI-powered solutions that meet their unique business needs.

Expanding SAP’s trusted AI model portfolio

is rooted in trust. Our customers expect and deserve AI that respects their data privacy, that it fits within their operational workflows, and that it understands the context and complexity of their industries. Cohere’s focus on security, efficiency, and enterprise applicability aligns perfectly with SAP’s approach to business AI and our generative AI hub.

Cohere Command models are lightweight, high-performing language models tailored for complex business tasks, with support for agentic workflows and multilingual operations. The Embed and Rerank models enable powerful enterprise search and retrieval capabilities, helping customers build accurate, context-aware RAG pipelines across structured and unstructured data.

Cohere models are designed to perform in production environments while respecting enterprise privacy requirements and compute constraints. And because Cohere shares our commitment to privacy-first design, these capabilities are built to serve even the most regulated industries, such as finance, healthcare, and the public sector.

SAP: Launch partner for Cohere’s reasoning model

As part of the partnership, 51·çÁ÷plans to be one of the first partners to offer Cohere’s upcoming reasoning model, a purpose-built, high-efficiency model designed to power agentic use cases.

We see enormous potential here. SAP’s vision for collaborative AI agents — capable of automating complex, multi-step tasks across systems — requires not just scale, but reasoning. Whether it’s helping consultants configure a system or enabling customer service to resolve cross-system issues, this next generation of AI requires models that can reason, plan, and act securely. Cohere’s reasoning model is built for exactly that.

We’re excited to partner with 51·çÁ÷and bring its enterprise customers the latest security-first models and solutions from Cohere. We’re especially excited that 51·çÁ÷will be one of the first partners to offer our upcoming reasoning model. 51·çÁ÷and Cohere share a vision for practical AI innovation, and our collaboration marks an exciting milestone as we unlock new efficiencies and growth for global enterprises.

Martin Kon, President and COO, Cohere

Powering real-world applications across industries

With this collaboration, 51·çÁ÷customers will be able to use Cohere models to solve pressing business challenges across industries, such as:

• Agentic task automation: Enable AI assistants that can take actions across enterprise tools and systems
• Multilingual RAG applications: Retrieve, rank, and summarize data from global policy manuals, compliance documents, or internal knowledge bases
• Secure document analysis: Understand long, structured, multimodal files like financial disclosures, M&A reports, technical manuals, or medical imaging
• Context-aware enterprise search: Improve search accuracy across unstructured content like emails, tables, or contracts

Customers will be able to easily access, test, and scale these models in production within SAP’s generative AI hub.

Expanding choice without compromise

With our partnership with Cohere, we are continuing to expand a growing ecosystem of AI capabilities that are open, secure, and business-ready. This partnership helps ensure that customers can choose the right model for their use case, while trusting that it meets SAP’s standards for quality, reliability, and compliance.

Together, 51·çÁ÷and Cohere are enabling enterprises to harness generative AI with confidence, whether they’re building knowledge assistants, automating processes, or delivering new intelligent services to users.

To learn more about our approach to enterprise-ready AI, visit .

Walter Sun is senior vice president and head of AI at SAP.

This work is complemented by two additional areas of progress: first, the expansion of Google Gemini models in SAP’s generative AI hub on 51·çÁ÷Business Technology Platform (51·çÁ÷BTP); second, the use of Google’s video and speech intelligence capabilities to support multimodal retrieval-augmented generation (RAG) for video-based learning and knowledge discovery in 51·çÁ÷products.

Together, these efforts reflect a shared commitment to deliver enterprise-ready AI that is open, flexible, and deeply grounded in business context.

Bringing AI agents together: laying the groundwork for interoperability

The future of work is agentic. Businesses are increasingly deploying AI agents that assist with real tasks — resolving customer issues, managing approvals, and collaborating across business functions.ĚýThis is why 51·çÁ÷is delivering a collaborative agent architecture with Joule to support cross-functional agentic workflows across 51·çÁ÷Business Suite.

But for these agents to deliver real value, they cannot operate within a single vendor landscape. They must be able to collaborate across various platforms, securely exchange information, and coordinate actions across complex enterprise workflows.Ěý This need for seamless interaction underscores why the represents a significant step beyond simple API integrations or enhanced tooling.

That’s why 51·çÁ÷has joined Google Cloud and other enterprise leaders as a founding contributor to the new A2A protocol. This open standard is designed to ensure agents from different vendors can interact, share context, and work together—enabling seamless automation across traditionally disconnected systems.

Consider a customer dispute resolution scenario: a representative receives a billing inquiry via Gmail. Instead of toggling between tools, they can invoke Joule directly from the email. Joule, acting as an agent orchestrator, initiates a dispute resolution process, engaging another Google agent that connects to Google BigQuery, where relevant transactional warehouse data resides. Together, the agents validate the issue, retrieve insights, and recommend a resolution — without manual system switching, data reconciliation, or context loss.

This is the kind of cross-platform collaboration the A2A protocol is designed to enable: AI agents working together to accelerate business outcomes, reduce friction, and enable people to focus on more strategic work. It also reinforces SAP’s vision for Joule as an agent orchestrator working across enterprise workflows: interoperable, proactive, and deeply connected to business context.

Expanding access to Google models in generative AI hub

Beyond agent interoperability, 51·çÁ÷is furthering its commitment to openness and flexibility by expanding access to Google models in the generative AI hub, a key capability of the AI Foundation on 51·çÁ÷BTP.

Through the generative AI hub, customers gain enterprise-grade access to a curated portfolio of leading foundation models. That portfolio now includes Google Gemini 2.0 Flash and Flash-lite, which join the existing support for Gemini 1.5 models already available through the hub.

This expanded model choice gives customers the flexibility to build and extend AI-driven solutions using high-performance, low-latency models optimized for enterprise workloads — while staying within SAP’s secure, business context-rich environment.

By combining Google’s model innovation with SAP’s deep understanding of enterprise processes, we enable customers to apply generative AI in ways that are not only powerful, but also practical, trustworthy, and fully aligned with how businesses operate.

Unlocking multimodal understanding with Google Video Intelligence

As part of our continued collaboration with Google Cloud, 51·çÁ÷is also advancing multimodal RAG, a highly requested capability among 51·çÁ÷customers, especially for video-based learning content.

Multimodal RAG enhances information retrieval and generation by integrating multiple data modalities — text, images, audio, and video — into a single, structured process. This approach enriches knowledge sourcing and elevates how users interact with training and support materials.

To address the complexity of extracting meaningful insights from video content, 51·çÁ÷leverages Google Video Intelligence for on-screen text detection across video frames, and Google’s Speech-to-Text API for accurate transcription of spoken audio. During the indexing process, these outputs are stored with corresponding timestamps, creating a structured foundation for retrieving relevant video segments with precision.

By grounding audio and visual content with time-aligned metadata, 51·çÁ÷enables users to search and retrieve specific, contextually relevant moments within a video, making the learning experience more intuitive, accessible, and impactful.

“As agentic AI evolves, seamless handling of multi-modal data — text, voice, enterprise videos, and images — becomes paramount,” said Miku Jha, director of AI/ML and Generative AI at Google Cloud. “This introduces significant challenges for agent interoperability. An open protocol like A2A is therefore indispensable, providing the necessary framework and flexibility for agents to effectively communicate and collaborate across these diverse modalities. Multi-modality is not simply a capability; it is a foundational requirement driving the next generation of interconnected agentic systems.”

This is another example of how 51·çÁ÷is integrating Google’s AI capabilities into business-relevant scenarios, helping customers unlock more value from their unstructured content and elevate the way knowledge is delivered across the enterprise.

Shared vision for business AI

These efforts reflect a broader strategic alignment between 51·çÁ÷and Google Cloud: a shared belief in AI that is open, composable, and grounded in real business context. Whether it’s shaping emerging standards for agent collaboration, providing choice through best-in-class models, or making unstructured content actionable, we are focused on helping our customers innovate with confidence — today and into the future.

To learn more about how 51·çÁ÷and Google Cloud are shaping the future of enterprise AI, visit and explore to see these innovations in action.

Walter Sun is senior vice president and head of AI at SAP.

With NVIDIA’s latest announcement of, 51·çÁ÷is strengthening its agentic AI strategy to help drive even greater business impact. These models allocate more compute time before responding to improve accuracy and enhance AI agents with advanced  decision-making and execution capabilities. By integrating them, Joule agents will become more adept at tackling complex business challenges—leveraging deeper contextual reasoning, making more precise decisions, and seamlessly interacting with enterprise data and systems to deliver more intelligent and autonomous operations.

Transforming business today: What 51·çÁ÷and NVIDIA have already achieved

The long-standing partnership between 51·çÁ÷and NVIDIA has already delivered AI innovations that are transforming business operations. Across industries, these AI innovations are revolutionizing how enterprises implement 51·çÁ÷solutions and develop applications on 51·çÁ÷Business Technology Platform (51·çÁ÷BTP).

For example, consultants working on digital transformation projects, such as those driven through RISE with SAP, are now leveraging 51·çÁ÷Joule for Consultants, enhanced with microservices, to quickly surface relevant insights from SAP-exclusive content. Consultants can ask questions in natural language to instantly retrieve precise guidance from past implementations and get assistance in interpreting ABAP code. This helps reduce the time spent navigating documentation and troubleshooting complex application logic, accelerating solution design, minimizing delays, and improving overall project efficiency.

For developers, AI-driven code generation has become a catalyst for innovation. Joule for developers, powered by microservices, part of software platform, helps generate ABAP code faster and more efficiently by enabling faster AI inferences for code generation tasks. 51·çÁ÷and NVIDIA collaborated on end-to-end model development for Joule for developers, including data processing and training the models. LLMs trained on ABAP and SAP’s enterprise logic enable developers to write, explain, and optimize SAP-specific code efficiently. As a result, enterprises can reduce development time, improve code quality, and accelerate innovation, modernizing their 51·çÁ÷environments more effectively.

Beyond implementation and development, AI is reshaping how businesses visualize and interact with their products. Leveraging technologies, 51·çÁ÷Intelligent Product Recommendation can bring real-time 3D visualizations to complex products, their offerings. Businesses can now simulate entire supply chains before making capital investments, optimize factory layouts with physics-based modeling, and enhance customer engagement through realistic, interactive product experiences that improve decision-making.

These AI-driven advancements are already reshaping operations in manufacturing, retail, and asset-intensive industries, enabling users to make more informed decisions, unlocking greater efficiency, and enhancing customer satisfaction.

What’s new: Advancing AI reasoning agents to transform work

“SAP’s AI agents, orchestrated by Joule, reason through challenges, solve complex problems, and drive efficiency,” said Kari Briski, VP, Generative AI Software for Enterprise, NVIDIA. “51·çÁ÷plans to integrate NVIDIA Llama Nemotron reasoning models to enhance AI-driven automation and enable businesses to optimize operations.”

With the integration of NVIDIA Llama Nemotron reasoning models, 51·çÁ÷will continue to advance the reasoning of Joule agents, redefining enterprise automation and allowing enterprises to scale automation with confidence and drive efficiency across end-to-end business processes.

The future of business AI

Together, 51·çÁ÷and NVIDIA are redefining how enterprises use and benefit from AI—bringing automation, intelligence, and efficiency to every facet of business operations. From advanced AI agents that execute complex cross-functional workflows to AI-driven development and immersive digital experiences, the possibilities are expanding rapidly.

about the 51·çÁ÷and NVIDIA collaboration today. Explore what AI can do for your .

Walter Sun is SVP and global head of AI at SAP.

But what exactly lies ahead? We’d like to share five key themes for AI in 2025 that undoubtedly come with challenges for businesses but also the potential to redefine what’s possible. Ready to glimpse into next year and beyond? Let’s dive in.

1.  Agentic AI: Goodbye Agent Washing, Welcome Multi-Agent Systems

AI agents are currently in their infancy. While many software vendors are releasing and labeling the first “AI agents” based on simple conversational document search, that will be able to plan, reason, use tools, collaborate with humans and other agents, and iteratively reflect on progress until they achieve their objective are on the horizon. The year 2025 will see them rapidly evolve and act more autonomously. More specifically, 2025 will see AI agents deployed more readily “under the hood,” driving complex agentic workflows.

Users will interact with a copilot for their tasks, which will deploy the request and coordinate among systems of multiple expert AI agents to complete more difficult tasks. Future AI agents, or , can collaborate to understand the business user, have all the context, and structure the problem to subsequently interact with these domain-specific expert AI agents — each performing specific sub-tasks that together complete a much more complex task. In the future, users will not even need to trigger an action. Instead, AI agents will proactively respond to business events such as incoming customer inquiries, supply chain disruptions, or demand surges. They will automatically prepare a decision workflow as far as they can before pinging the human user for feedback.

If we look at a five-year horizon, AI agents will simplify significant portions of workflows, even aspects that have been resistant to automation, such as exceptions in customer service, long-tail administrative tasks, and specific programming activities like coding or debugging software. AI agents will be flexible and can plan, fail, and try something else or self-correct based on reasoning. AI agents will handle and complete routine, repetitive tasks end-to-end as effectively and often even more effectively than humans, leading to increased productivity and demonstrable cost savings. Agents will be more adaptable and robust than conventional robotic process automation (RPA) for longtail and highly extensive tasks. This means figuring out the best result out of many possible outcomes, which is almost impossible to hardcode in an RPA algorithm with classical automation methods.

Adopting AI in these domains will also shift workforce dynamics, with human roles evolving to focus on anticipating uncommon scenarios, coping with ambiguity, factoring in human behavior, making strategic decisions, and driving genuine innovation — complemented, not replaced, by AI capabilities. 

In short, AI will handle mundane, high-volume tasks while the value of human judgement, creativity, and quality outcomes will increase.

2. Models: No Context, No Value

Large language models (LLMs) will continue to become a commodity for vanilla generative AI tasks, a trend that has already started. LLMs are drawing on an increasingly tapped pool of public data scraped from the internet. This will only worsen, and companies must learn to adapt their models to unique, content-rich data sources. Model improvements in the future won’t come from brute force and more data; they will come from better data quality, more context, and the refinement of underlying techniques. Companies must spend more time innovating to make better models through fine-tuning and model adaptation rather than just training larger and larger models. Neurosymbolic AI techniques, especially knowledge graph, will see a renaissance since they can provide both learning objectives for foundation models and context to significantly improve the performance of generative AI while reducing hallucinations.

We will also see a greater variety of foundation models that fulfill different purposes. Take, for example, physics-informed neural networks (PINNs), which generate outcomes based on predictions grounded in physical reality or robotics. PINNs are set to gain more importance in the job market because they will enable autonomous robots to navigate and execute tasks in the real world, from warehouses to manufacturing plants, or models trained on tabular, structured data, like 51·çÁ÷Foundation Model, and can handle tasks that LLMs cannot do well, like predictions of numeric values.

Models will increasingly become more multimodal, meaning an AI system can process information from various input types. AI applications will eventually evolve into “any-to-any” modality solutions capable of understanding, processing, and reasoning across text, voice, image, video, and sensor data within a single model. In addition, smaller and more specialized LLMs with scalable finetuning techniques and the ability to work on any device will become more common, a trend that may lead to hyper-personalized models for organizations or even individuals in the future.

Enterprises will shift toward strategies utilizing multiple foundation models (not to be confounded with multimodal capabilities in a single model, described above), leveraging a diverse set of AI models and techniques tailored to specific use cases. This is backed by the trend of fine-tuning small slices of models, which requires fewer resources and much less data, resulting in full model flexibility and enabling businesses to extract more value from their unique data and gain a competitive edge. Enterprise software vendors will offer or extend integrated AI model marketplaces and platforms that support seamless model deployment, management, and updating. Benchmarking and lowering model switching costs will help deploy the same use cases in heterogeneous environments. Context equals value. Knowledge graph technology has been around for 40 years and is now seeing a revival because it can overcome key LLM challenges, such as understanding complex formats, hierarchy, and relationships between business data. Knowledge graphs offer data meaning and explain the relationship between entities, significantly supercharging the abilities of LLMs. The next step in this journey will be large graph models, allowing further advancement in generative AI.

Implicit knowledge is power, and making knowledge explicit to others is a superpower.

3. Adoption: From Buzz to Business

While 2024 was all about introducing AI use cases and their value for organizations and individuals alike, 2025 will see the industry’s unprecedented adoption of AI specifically for businesses. More people will understand when and how to use AI, and the technology will mature to the point where it can deal with critical business issues such as managing multi-national complexities. Many companies will also gain practical experience working for the first time through issues like AI-specific legal and data privacy terms (compared to when companies started moving to the cloud 10 years ago), building the foundation for applying the technology to business processes.

From a technological perspective, while 2024 saw significant advancements in AI, 2025 will see companies focus on making these advancements more meaningful through seamless data integration, ultimately enhancing the accuracy and significance of AI-powered outcomes and boosting adoption. Lastly, in 2025, we might glimpse a shift in the software business model from building static software features and functions to an outcome-as-a-service model focused on achieving process objectives.

4. User Experience: AI Is Becoming the New UI

AI’s next frontier is seamlessly unifying people, data, and processes to amplify business outcomes. In 2025, we will see increased adoption of AI across the workforce as people discover the benefits of humans plus AI.

This means disrupting the classical user experience from system-led interactions to intent-based, people-led conversations with AI acting in the background. AI copilots will become the new UI for engaging with a system, making software more accessible and easier for people. AI won’t be limited to one app; it might even replace them one day. With AI, frontend, backend, browser, and apps are blurring. This is like giving your AI “arms, legs, and eyes.” While power users will still have singular, expert interfaces, most users will demand flexibility across multiple access patterns. At the same time, there will be a growing acceptance of longer inference times for high-quality answers to complex, previously unsolvable problems and actions in domains requiring deep analysis and research. Ultimately, users will recognize the trade-off between latency and complexity of tasks handled by AI.

Importantly, we will see organizations move beyond viewing AI as a collection of productivity tools and begin reimagining their workforce as a network of collaborative intelligence with AI agents and humans working to accelerate innovation within the enterprise. For example, combining human expertise in strategic thinking with AI’s strengths in large-scale analysis and pattern recognition will create new competitive advantages for companies that effectively orchestrate these hybrid intelligence networks to drive breakthrough discoveries and market opportunities. Next year will also mark the early stages of a significant shift in how humans and AI work together, with agents evolving into workflow partners, taking initial steps toward independently navigating software environments and automating routine tasks – from data analysis and report generation to schedule coordination and software testing. This will also start a longer journey toward transformed work processes and patterns, with forward-thinking organizations developing new roles, metrics, and training approaches for effective human-AI task collaboration.

5. Regulation: Innovate, Then Regulate

It’s fair to say that governments worldwide are struggling to keep pace with the rapid advancements in AI technology and to develop meaningful regulatory frameworks that set appropriate guardrails for AI without compromising innovation. The regulatory landscape will become even more fragmented, with the tracking hundreds of AI regulations under discussion worldwide. This requires evaluating model compliance with and technical interpretation of various regulatory frameworks.

In 2025, the discussion will shift from what we try to regulate from a technical standpoint to how we innovate and what we deem fundamentally human. This discussion will elevate the role of humans, contribute a much more positive perspective, and help shape a long-term vision for how we want humanity and AI to live and work together. 

In this environment, it will continue to be critical for companies developing and deploying AI technology to adhere to responsible principles around safety, security, and ethical use. This will also help set the stage for important precedents and compliance.

Executing on the Themes in 2025

Indeed, these are just a few of what we are sure will be many exciting advancements for AI in 2025. Overall, the biggest takeaway from the year ahead will be making existing breakthrough technology more meaningful. We will see AI much deeper and almost invisibly embedded in consumer and enterprise applications and witness more advancements in how vendors and organizations that use these applications embed their individual contexts and data into AI seamlessly.

Getting to the point of leveraging AI generally, however, will require businesses to take advantage of a modern cloud suite with unified data access and harmonized data models to overcome data silos and fully benefit from AI innovation that spans across the whole enterprise. This will drastically increase the accuracy and significance of AI-powered outcomes, ultimately boosting adoption, specifically in the enterprise space.

We can’t wait to see what the future holds.

Sean Kask is vice president and head of AI Strategy for 51·çÁ÷Business AI at SAP.
Walter Sun is senior vice president and global head of AI for 51·çÁ÷Business AI at SAP.
Jonathan von Rueden is head of AI Frontrunner Innovation for 51·çÁ÷Business AI at SAP.

The generative AI hub enables enterprise customers to easily access a broad range of commercial and open-source large language models (LLMs) in a safe and secure environment. This milestone marks a new chapter in our long-standing partnership with AWS, combining SAP’s AI innovations and enterprise expertise with Amazon’s latest and most advanced AI capabilities and technology solutions to unlock powerful opportunities for businesses.

Amazon Nova joins a growing portfolio of top-tier commercial and open-source models available through the generative AI hub. By making the new Amazon Nova models publicly available simultaneously with their release on AWS, 51·çÁ÷helps ensure that customers can immediately leverage the latest innovations to build AI-driven solutions that harness the full business context in 51·çÁ÷data.

Tailored AI Solutions for Every Need

The Amazon Nova release includes three state-of-the-art understanding models that are available today, enabling businesses to choose the right model for their unique needs:

• Amazon Nova Micro: Text-only model that delivers the lowest latency responses at very low cost
• Amazon Nova Lite: Very low-cost multimodal model that is lightning fast for processing image, video, and text inputs
• Amazon Nova Pro: Highly capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks

With these models now accessible through the generative AI hub, developers can efficiently leverage them alongside the extensive capabilities of 51·çÁ÷Business Technology Platform (51·çÁ÷BTP) to build and scale AI-powered solutions that seamlessly work with 51·çÁ÷applications.

Streamlined Access and Integration

SAP’s generative AI hub helps simplify how businesses access and deploy advanced AI models.ĚýThe Amazon Nova models, alongside other leading offerings, are available through a unified interface deeply integrated with 51·çÁ÷BTP.ĚýThis helps ensure customers can automate workflows with precision, build multimodal AI solutions tailored to specific challenges, and fully utilize the business context embedded in 51·çÁ÷data.

By combining Amazon Nova models with 51·çÁ÷BTP services like workflow automation and vector databases, businesses can develop customized AI solutions that can scale effortlessly.ĚýFor example, developers can create new skills for Joule, SAP’s AI copilot, leveraging Amazon Nova models and 51·çÁ÷HANA’s vector capabilities to process enterprise-specific data.ĚýThis helps ensure AI outputs are not only accurate and relevant but also actionable, driving measurable results across operations.

51·çÁ÷and AWS: Unlocking the Future of Business AI

The availability of Amazon Nova models on the generative AI hub marks a significant milestone in SAP’s long-standing collaboration with AWS. Together, we continue to address critical enterprise challenges such as data security, regulatory compliance, and scalability, ensuring businesses can confidently adopt AI solutions that deliver tangible value. This partnership underscores our shared vision of making advanced AI accessible and impactful for businesses worldwide, unlocking new levels of efficiency and innovation.

With access to Amazon Nova Micro, Amazon Nova Lite, and Amazon Nova Pro, 51·çÁ÷customers can reimagine their operations — automating workflows, enhancing insights, and driving complex decisions. We’re excited to see how our customers and partners leverage these capabilities to achieve impactful business outcomes.

Learn more about innovating with and the new .

Grounded in an AI ethics framework, 51·çÁ÷collaborates with partners to create insights from industry-specific data and deep process knowledge, driving exponential value for businesses. By fostering a multi-partner approach, 51·çÁ÷ensures flexibility and prevents vendor lock-in, empowering customers and partners to navigate the AI landscape effectively. 

As a next step in that journey, 51·çÁ÷is integrating the Meta Llama 2 and Meta Llama 3 models into the generative AI hub to enable customers to create dashboards based on rich content with Llama qualitative conversational outputs, and to explore use cases built with Meta Llama 3.  

The provides an overview of SAP’s business AI strategy, highlighting the unparalleled opportunity for the 51·çÁ÷ecosystem to build an array of solutions with generative AI capabilities and extensions on 51·çÁ÷BTP.Ěý

51·çÁ÷Analytics Cloud AI Portfolio Expansion 

51·çÁ÷will make Joule available in 51·çÁ÷Analytics Cloud to let planning and analytics users get work done faster and drive better business outcomes in a secure and compliant way. The wide range of capabilities offered by Meta Llama 3 will enable Joule users to get assistance with key activities, including the auto-generation of custom scripts to extend dashboards, as well as delivery of the most accurate scripts directly in 51·çÁ÷Analytics Cloud.Ěý 

The goal is to integrate generative AI into major workflows of 51·çÁ÷Analytics Cloud. Transforming these workflows will unlock innovative possibilities for distinct planning and analytics roles that are common in organizations globally. More information on how 51·çÁ÷Analytics Cloud with Joule will help planning and analytics users get work done more efficiently .Ěý

Availability of Meta Llama 3 on Generative AI Hub 

51·çÁ÷will make Llama 3 available in Q2 2024 in the generative AI hub in 51·çÁ÷AI Core given customer interest in Llama 3. The generative AI hub makes it simple to build generative AI use cases for 51·çÁ÷applications, thanks to its value-adding features around 51·çÁ÷integration and the ability to orchestrate language model interactions.

Adding Meta Llama 3 to the generative AI hub allows developers in the 51·çÁ÷ecosystem to leverage its value-adding features. The 70B variant of Meta Llama 3 will be the largest open-source model available on the generative AI hub.Ěý

51·çÁ÷SuccessFactors Text-to-Image Generation Requirements 

51·çÁ÷customers are exploring ways to enhance the visual appeal and relevance of images within the 51·çÁ÷SuccessFactors HCM suite. The current process of curating and uploading images can be time-consuming and costly, often resulting in generic images that only partially cater to specific scenarios while also incurring expenses associated with stock image licensing.

Integrating the latest iteration of Meta’s open-source large language model (LLM) offers numerous benefits. Creative features from models such as Llama 3 come into play in a scenario like this, empowering users to generate custom images that better meet their needs within 51·çÁ÷SuccessFactors software. Whether it is customizable course content thumbnails and banners for learning, tailored images for assignments in 51·çÁ÷SuccessFactors Opportunity Marketplace, unique images for different themes on the home page, or personalized profile and banner images on people profile, the impact will be far reaching.Ěý

By streamlining the image creation process, reducing costs, and ensuring that every image is suited to its context, this feature will undoubtedly improve user engagement and the overall platform experience. 51·çÁ÷customers should get ready to explore a new world of custom imagery tailored to their exact specifications, sparkling excitement and intrigue among stakeholders.

Walter Sun is senior vice president and global head of AI at 51·çÁ÷SE.

This collaboration is more than just a meeting of minds; it’s a symbiotic combination of AI expertise and technology that opens a world of possibilities for 51·çÁ÷customers.

Mistral AI’s success in developing advanced LLMs, including its renowned open-weight models Mixtral 8x7B and Mixtral 8x22B, and more expansive enterprise-grade “Large” model, is set to complement the 51·çÁ÷suite of AI-enabled solutions. The collaboration will enable direct accessibility to Mistral AI’s models through 51·çÁ÷or through 51·çÁ÷Business Technology Platform (51·çÁ÷BTP) applications with generative AI capabilities.

What does this mean for 51·çÁ÷customers? Simply put, it’s about empowerment and access to AI from a European LLM provider. Access to Mistral AI’s latest models through the generative AI hub in 51·çÁ÷AI Core will enable 51·çÁ÷customers to enhance productivity, streamline their operations, and accelerate their digital transformation journey.

Whether through integrating AI with 51·çÁ÷BTP or developing bespoke solutions through direct access to Mistral AI LLMs, the potential for innovation is limitless.Ěý

“We are excited about entering a partnership with Mistral AI and making the company’s LLM accessible to both our developers and our customers through the generative AI hub in 51·çÁ÷AI Core on 51·çÁ÷BTP,” said Philipp Herzig, chief AI officer of 51·çÁ÷SE. “Together, we can truly make a difference by building AI-enabled solutions that create immediate value for users, organizations, and entire industries. We are particularly proud that two European technology companies are collaborating on bringing AI forward.”

“We are pleased to embark on this partnership with SAP,” said Arthur Mensch, CEO of Mistral AI.Ěý“We foresee the new horizons this collaboration will open up, enabling us to further our mission of making AI accessible to all. We are looking forward to witnessing the potential of our AI models to support innovation and streamline operations for SAP’s customers.”

The ambitions don’t stop there: 51·çÁ÷and Mistral AI are committed to exploring new applications of AI across various industries. By leveraging the combined strengths, this is not just about driving innovation, but about creating new business opportunities and delivering tangible value to 51·çÁ÷customers.Ěý

Stay tuned as we begin this exciting journey together. The future of enterprise software is bright, and with partners like Mistral AI, we are ready to illuminate the path forward.Ěý

Walter Sun is senior vice president and global head of AI at 51·çÁ÷SE.

With Gartner finding more than 80% of enterprises using generative AI application programming interfaces (APIs) or models, and/or deploying generative AI-enabled applications in production environments by 2026, we know there is vast opportunity ahead of us to build the tools and layout the pathways to enable every developer to make an impact.

We believe AI will have a profound effect on the way engineers and developers work in three fundamental ways. 

• First, how we build our software and applications
• Second, what we build into the DevOps cycle tools and platform
• And finally, who can code using natural language conversations, extending programming beyond low coders to no coders

Development will become even more democratized to anyone with great ideas and creativity.

At SAP, we are at the forefront of business AI. SAP Business AI is about achieving real business results for the enterprise and making AI relevant, reliable, and responsible.

We have embedded AI capabilities into our solutions including Joule, our new generative AI copilot, which understands your business. 

We are working relentlessly to bring these generative AI use cases to our customers, including educating 51·çÁ÷employees on best practices and providing AI tools internally to accelerate generative AI-embedded product development. We are doing the same for developers. Here are three specific ways 51·çÁ÷is empowering every developer to become an AI developer.

Joule: Natural Language Programming

, our generative AI copilot, revolutionizes the way users interact with 51·çÁ÷business systems and data. With Joule, users can easily navigate across 51·çÁ÷applications, find information, execute transactions, collaborate with colleagues, and receive proactive recommendations.  

Instead of conducting laborious manual searches, Joule users can reduce their average search and query time by up to 80% by asking natural language questions and receiving intelligent answers on the wealth of business data and insights available across the 51·çÁ÷portfolio.

51·çÁ÷Build Code

Our new application development solution, , draws on the power of Joule to create data models, application logic, and test scripts. 51·çÁ÷Build Code provides a simplified developer experience that drives productivity and supports fusion development by teams of professional and citizen developers.

AI Foundation on 51·çÁ÷Business Technology Platform

We believe that business AI capabilities need to be directly embedded into business applications and extensions. The  is a one-stop shop for developers to do exactly that, providing ready-to-use AI services and tools to accelerate the development of generative AI-infused applications, in a secure and trusted way. 

AI Foundation includes everything developers need to run their business-ready AI in the applications they have built on 51·çÁ÷BTP, from ready-to-use AI services to AI runtime and lifecycle management. We provide the tooling for the management of generative AI capabilities and ensure business data connectivity. We designed it with security, governance, and trust from the ground up. 

Looking Forward: The Right Model for the Right Business Need

In early 2024, the AI Foundation will also include a generative AI hub, giving developers instant access to a broad range of large language models (LLMs) from different providers, such as OpenAI’s ChatGPT or Falcon-40b. We will soon add more models.

With this, developers can experiment by submitting a prompt to multiple LLMs to compare the generated outcomes and identify the best model for the task. They can then power mission-critical processes with complete control and transparency with features like built-in prompt history, all delivered with enterprise-grade security and data privacy. 

Our generative AI hub will also provide tooling for developers to supercharge their generative AI development processes and access to tools and techniques to extract the most relevant information from business and other data.

To get the most value in traditional business areas such as finance, sales, or supply chain, some enterprise customers require foundation models designed specifically for business context. 51·çÁ÷is committed to fine-tuning generic LLMs on 51·çÁ÷anonymized data and, in the longer term, to creating proprietary foundation models based on our vast structured business data. 

These models will be able to address questions that users face every day in business that LLMs cannot, such as predicting invoice payment dates and supplier delivery quality or proposing efficiencies to a business process. 

At SAP, we are building the tools that will enable every developer to become a business AI developer, and in the process to become the true architects of the future.

Walter Sun is global head of AI at SAP.