What is Agentic AI?

Agentic AI refers to AI systems that pursue a given goal independently across multiple steps: they plan, choose tools and data sources, carry out actions, check the result, and adjust their approach. It is usually built on a large language model that is combined with tool access, memory, and a control loop to form one or more AI agents. What matters is not that the AI generates text but that it acts.

Why does this matter?

With agentic AI, AI shifts from answering to getting things done – and therefore from the single request into the running business process. That opens up automation where fixed rules used to fail, but raises new questions: what mandate does the agent have, what data does it decide on, and how can what it did be evidenced afterwards? Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, mainly due to unclear business value and inadequate risk controls.

What does this look like in practice?

A generative AI writes a payment reminder when asked. An agentic system is given the goal "reduce receivables older than 60 days": it pulls the affected invoices from the ERP, checks per case whether a complaint is open, drafts staged reminders, routes cases with more than €10,000 in dispute to a case handler, and measures two weeks later how much has come in.

Not to be confused with

Not the same as generative AI, which produces content on request; agentic AI uses such models to carry out tasks. Nor the same as classic automation, where every step is fixed in advance, whereas an agentic system decides at runtime which step makes sense next.