Agentic AI and AI Agents: How autonomous systems are transforming businesses and why shadow AI is becoming a key risk

Agentic AI, or "acting artificial intelligence", marks a turning point in the digital transformation. While previous AI systems were designed to react to predefined inputs, today's agents can independently plan, make decisions, and act. According to "Gartner" and their so-called "AI Hype Cycle", AI agents are currently among the technologies with the highest development potential (Khandabattu, 2025).

What are AI agents and how have they evolved?

At the beginning of this decade, AI agents were primarily used as extensions of large language models. Systems such as AutonomAgent or Claude Code could automate simple tasks, such as data research or software testing. Today, such systems operate in complex processes: they analyse data streams, make business decisions, and interact with other applications in real time. Companies like the Austrian energy enterprise OMV are already using agent-based chatbots to provide regulatory information more quickly. According to the digital transformation expert magazine, Digital Business Magazin, this reduced processing time by 20 percent and training costs by 25-fold. (Ingerfeld, 2025)

BCG and MIT describe in a joint study that 35 percent of companies worldwide are already relying on agentic AI, and another 44 percent plan to do so in the coming years. Particularly remarkable is the shift in perception: 76 percent of executives now see agentic AI as a "colleague" rather than a tool. (Ransbotham et al., 2025) Provided they are managed correctly, these systems combine human decision-making ability with machine speed, thereby enabling a new level of efficiency.

Gartner describes AI agents in its "Emerging Tech Impact Radar 2025" as one of the technologies that will fundamentally change the interaction between humans and machines (Nguyen & Casey, 2025). Together with intelligent reasoning models, they will ultimately revolutionise how we interact with technology. Reasoning models are AI systems that not only process data but also draw logical conclusions and analyse problems much like a human would. Through this ability for adaptive learning, agents accelerate workflows across almost all industries.

What are AI agents used for?

In software development, customer service, and drug discovery, specialised AI agents are already delivering productivity and time-to-market increases of 50% and more. AI agents developed for tax processes now generate complex documents within a single day that previously took up to two weeks. And across the financial sector as a whole, AI agents are fundamentally changing how data is collected and analysed, whilst simultaneously improving the speed and quality of audits. (Abbatiello & Priest, 2025)

According to the website aiagentslist.com and its overview of models in the field of AI agents, there are currently at least 626 AI agents across 19 categories. These include areas such as Customer Service, Marketing, Finance, Coding, Content Creation, Digital Workers, and Business Intelligence (aiagentslist.com, 2025). This shows that AI is already widely deployed today.

The largest areas of application for Agentic AI lie in the automation of complex business processes, data-driven decision support, and workflow optimisation. Experienced organisations with high agentic usage are currently investing primarily in using AI as an analytical partner to strengthen human oversight rather than as an independent decision-maker. 79% invest in AI-powered insights for decision-makers, whilst fully autonomous "AI decides and implements" scenarios are significantly less common at 54%. (Ransbotham et al., 2025)

What does Agentic AI bring with it?

However, with this widespread adoption, a phenomenon known as "shadow AI" is growing. This is the uncontrolled use of AI systems outside of central IT governance. By 2028, 33% of all enterprise applications will contain Agentic AI, up from less than 1% in 2024. These autonomous systems can make independent decisions, learn, and continuously improve (CyberArk, 2025).

It becomes particularly critical when agents independently access email systems or customer data. Here, a seemingly harmlessly configured mail agent, for example for automatically answering enquiries, can unintentionally share sensitive data or perform unauthorised actions.

Gartner sees AI data privacy as the next crucial development field for AI technologies: "The complexity of AI agents makes them vulnerable to data security and governance issues. Without human oversight, organisations lack real trust" (Khandabattu, 2025).

Since anyone can create AI agents, uncontrolled shadow AI structures are emerging in enterprises. Employees often implement agents without IT approval, leaving organisations blind to their own security risks. Already today, machine identities outnumber human identities by a ratio of 82:1. Because AI agents require access to sensitive systems but lack human security awareness, compromised agents with extensive privileges can cause significant damage. A lack of regulations further amplifies these risks. (CyberArk, 2025)

Regulatory pressure is also mounting: the EU AI Act envisages that many autonomous agents will be classified as high-risk systems. This means extensive documentation requirements, human oversight, and the obligation to ensure the traceability of decisions (European Commission, 2025).

What will the future with AI agents look like?

In the future, AI agents will not only automate individual tasks but will orchestrate entire business processes. Between 2026 and 2028, market analysts expect the second developmental phase of so-called multi-agent systems, which communicate with each other and develop joint strategies (Deloitte, 2025).

Agentic AI is not simply smarter automation. It provides companies with a strategic roadmap to efficiently reduce costs, drive revenue growth, and unlock the full potential of their employees. Although autonomous systems bring major benefits, their complexity requires a pragmatic and aligned development of strategy, technology, data, workforce, governance, and change management. Employee roles will shift significantly: from monitoring and feedback in the early stages to orchestration and strategic control once agentic AI is more deeply integrated. This makes dynamic talent development and a thoughtful redesign of work essential. Companies that act early and view autonomy as a phased transformation process secure sustainable advantages and shape the standards of their industry. (Deloitte, 2025)

For European companies, the step into the world of Agentic AI should be carefully planned. Anyone wishing to seize the opportunities must make governance, data protection, and transparency a priority. After all, only those who control their agents will prevent productive AI from working in the shadows.

At NENNA.AI, we are convinced that AI should not be blocked, but enabled in a sensible and secure way. This is precisely where we come in. With our AI Privacy Platform, we enable companies to leverage the full potential of Artificial Intelligence without compromising on data protection and security.

Sources:

Ingerfeld, M. (2025, 23 April). AI Agent Systems: Wie sie die Geschäftsabwicklung verändern. Digital Business Magazin. https://www.digitalbusiness-magazin.de/ai-agent-systems-wie-sie-die-geschaeftsabwicklung-veraendern-a-f4fcaf2b5c6327f823279b6b35b04b80/

Ransbotham, S., Kiron, D., Khodabandeh, S., Iyer, S., & Das, A. (2025). The Emerging Agentic Enterprise: How Leaders Must Navigate a New Age of AI. MIT Sloan Management Review & Boston Consulting Group. https://sloanreview.mit.edu/projects/the-emerging-agentic-enterprise-how-leaders-must-navigate-a-new-age-of-ai/

Nguyen, T., & Casey, D. (2025, 14 February). Emerging Tech Impact Radar: Generative AI. Gartner. https://www.gartner.com/en/documents/6179023

Abbatiello, A., & Priest, D. (2025). AI agents are the future of work. PwC. https://www.pwc.com/us/en/tech-effect/ai-analytics/ai-agents.html

aiagentslist.com. (n.d.). AI Agents List / AI Landscape 2025: Complete visual map of AI tools and agents. https://aiagentslist.com/ai-agents-map

Khandabattu, H. (2025, 8 July). Der Hype Cycle 2025 für KI geht über GenAI hinaus. Gartner. https://www.gartner.de/de/artikel/hype-cycle-fuer-kuenstliche-intelligenz

CyberArk. (2025, 10 October). The business rewards and identity risks of agentic AI. Harvard Business Review. https://hbr.org/sponsored/2025/10/the-business-rewards-and-identity-risks-of-agentic-ai

European Commission. (2025). Regulatory Report: EU AI Act and AI Agents.

Deloitte. (2025, September). Agentic enterprise 2028: A blueprint for cost savings, job creation, and faster growth through agentic AI. Deloitte. https://www.deloitte.com/content/dam/assets-zone3/us/en/docs/services/consulting/2025/agentic-ai-enterprise-2028.pdf