AI agents – what they are (and what they are not)

In recent posts, I have written about why many AI initiatives stop before they create real value - and why data, context and RAG are prerequisites for mature AI use. Now it's time to clear up a term that keeps popping up in the AI ​​debate: AI agents.

Let's start precisely. In Generative AI for Business, Shubin Yu describes agents as systems that not only respond to requests, but can plan, execute and follow up tasks over time - within defined frameworks. An agent has a clear goal, access to relevant data and tools, ability to make sequential decisions, and clear boundaries for responsibility and control. It also says a lot about what an agent is not: it is not just ChatGPT in a new wrapper, it is not automation without context, and it is not "autonomous" without governance. Agents are workflows with intelligence – not magic.

A commercial example that many will recognize is sales. Without agents, salespeople use AI to write emails, summarize meetings and get suggestions for offer texts. Useful, yes – but still individual and fragmented. With an agent, the picture can look completely different. A sales agent can follow the entire pipeline automatically, analyze CRM data, previous deals and customer behaviour, suggest the next best course of action per customer, notify when risk in a deal increases and prepare management decisions before forecast meetings. Here we have moved from AI as a tool to AI as operational support in a core process. The value is not in the text, but in the decisions that are improved.

That is also why this is primarily a leadership theme, not an IT project. Most organizations are not ready for agents, even if the technology exists. McKinsey describes in 2026 how they themselves use tens of thousands of AI agents internally, with an ambition that each consultant should have at least one agent to support the work. Not because the agents are "smart", but because the processes are clear, the ownership is defined and the decision-making space is clarified. Agents require maturity in governance, not just maturity in technology.

There is also an important notice to managers here. The more autonomous AI becomes, the more important the answers to some fundamental questions become: Which decisions can be delegated – and which cannot? Who owns the consequences when something goes wrong? And how do we stop an agent – ​​and when? Agents reinforce the organization as it is. If the structure is unclear, the ambiguity is amplified.