SAP Forecast: 5 AI Trends in 2026

SAP sees five defining trends related to AI, all of which show that AI will continue to shape corporate operations in the coming year, penetrating ever deeper into business decision-making and corporate operations.

“In the future, data governance and responsible AI will not be a competitive advantage, but a threshold for entry in the modern business world. The effectiveness of AI depends on quality and connected data. Isolated data islands limit performance, so investing in modern cloud-based systems and harmonizing data is essential”

– emphasized Péter Hidvégi, Managing Director of SAP Hungary, in connection with the forecast.

1. “Specialized” base models will appear, going beyond the limitations of general LLMs

The first big wave of generative AI was brought by general, “good for a little bit of everything” language models (like ChatGPT), which were trained on internet-scale data. These are really useful in many corporate tasks – for example, document summarization, copywriting or coding assistance. However, their limitations are becoming increasingly apparent: they are only limited in their ability to perform accurate calculations and reliable predictions, they do not have “inside” knowledge of a company’s operations and data, and they are not suitable for tasks performed in real, physical environments (such as robotics) where mistakes have immediate consequences.

The next big step in the corporate world will therefore be the emergence of specialized, task-specific basic models. These also learn from a lot of data, but they do not want everything at once, but focus on one area and data type. By 2026, these target models will be able to work in an increasing number of complex business tasks: from video and physical simulations to supporting industrial processes and robots. Of particular promise are models that are “native” to structured data (tables, sensor data, transactions) and can quickly and accurately predict, identify outliers (anomalies), optimize processes, and provide tangible decision support in finance, manufacturing, logistics, or the supply chain.

Overall, these specialized models are often faster, more accurate, and cheaper to run than their general-purpose counterparts. As a result, companies can deploy working predictive or optimization solutions in days, not months—and thus AI can become a major driver of business intelligence and automation.

2. The era of autonomous AI agents has arrived

The development of enterprise software could reach a new level by 2026: AI will not be just a “smart extra”, but a core element of operations. They can predict problems, make suggestions, and even automatically complete certain processes, while adhering to corporate rules and approval circles.

This AI-centric architecture is built on intelligent agents: “digital colleagues” who communicate in natural language, manage system-wide processes, and react to changes. The spread of agents could bring a big leap, especially in administrative areas (HR, finance, procurement): they will not automate individual tasks, but entire process chains. The point is not to “replace” the expert, but to increase their scope: the agent collects, prepares, recommends – the decision and responsibility remain with the expert. This increases productivity, but roles are transformed.

3. Regulation of AI agents becomes a key issue

As more and more autonomous AI agents are introduced into corporate systems, companies must treat them in a new way – as if they were digital employees. These agents do not just perform a single step, but complex, multi-step tasks: for example, they process documents, prepare decisions, or even complete a complete travel planning process. These and similar AI agent functions go beyond classic automation.

Due to the rapid spread, companies need to develop rules and frameworks that cover the “lifecycle” of agents (deployment, update, shutdown), transparency (what they did and why), compliance with company rules and regulations, the order of human-AI cooperation (approval practices), and continuous measurement and control of their performance.