What are AI Agents?

The 'agentification' of work is more than only productivity.

2 Frameworks to understand what AI Agents are

2025 marks the tipping point in artificial intelligence history, we moved beyond the 'co-pilot' era and entered the age of autonomous AI Agents. Business adaption of agents has tripled between Q1 and Q2 and 90% of companies have passed the experimental phase (source). This shows that agents are not something futuristic but are becoming the backbone of the modern workflow.

An AI Agent is essentially a piece of software that perceives, decides and executes what is needed to achieve a goal. But there is a big distinction between AI Assistants and AI Agents: assistants are AI's that we use to do things, while AI Agents are doing things for us.

Yet amid the excitement over this 'agentification of everything', confusion remains. What makes one agent suitable for financial analysis and another for customer service? To truly understand how to define your agent-strategy, you need to look at these systems through two complementary lenses: the functional framework (how they think) and the focus framework (what they do).

First view: the functional hierarchy (how an agent thinks)

This classic AI framework classifies agents based on their internal architecture, intelligence and memory. This is the base for their capabilities and shortcomings.

1. The basic Reflex agent (pure reactive):

This is the most basic type, working on the 'action-reaction' principle (or the 'if-then' principle) and only looks at what is happening at this moment. It has no history, no memory, or context ).

Think about a basic email-rule: if a mail arrives with 'invoice' in the subject line, send out a basic reply. These agents are fast and reliable in structured and predictable tasks. But they fail when context or history is needed.

2. The model-based Reflex agent (the agent who remembers)

This agent adds a very crucial layer: an internal model of its environment (the 'state'). It remembers what's happening in this environment and knows how its own action can affect the environment.

This memory enables the agent to predict its own actions and to infer what is happening in unseen parts of its environment.

A good example of this is a network monitoring system that combines log data and metadata to understand the overall network condition (the 'state')

3. The Goal based agent (the agent that plans)

This agent moves beyond reacting—it plans. Instead of following rules, it works toward an explicit goal (a desired future state). The agent simulates different sequences of actions to figure out how to reach that goal.

Example: An inventory management system that plans reorder schedules to maintain target stock levels.