AI agents and technology
What is an AI agent? Definition and examples
Also: Agent, agentic AI
In short
An AI agent is a program that uses a language model to pursue a goal, plan its own steps and use tools to get there. Unlike a chat, it does not only answer, it acts. How well it works depends on its context and on clear approvals.
Why it matters: AI agents take over whole work steps, so they need context and limits.
- 01
Goal
An assignment with a clear result
- 02
Plan
Breaks the task into steps
- 03
Act
Reads sources, uses tools
- 04
Check
Compares result and goal
- 05
Approve
A person decides
Why does AI agent matter for your company?
A chat assistant answers a question. An AI agent completes a task: it breaks it into steps, gathers information and checks its interim result. Work shifts from people who type to people who give assignments and approve results.
For leadership one thing matters most: an agent is only as good as the context it reads and the rights it holds. If it does not know the current decision, it carries out the old one at full speed. That is why every agent needs a valid state, clear limits and approval for anything with consequences.
Example: An agent prepares offers
A sales team has an AI agent prepare offers. It reads the CRM, the price list and past offers and writes a draft. After the decision “Strategy 2027: focus on the DACH mid-market, no US expansion” it still proposes US packages, because the old template is still in its sources.
The fix is not a better model. The team gives the agent the approved decision as context, removes the old template from its sources and has a person check every offer before it goes out. From then on the agent works from the state leadership decided.
3 common mistakes to avoid
Starting an agent without context
An agent without current context guesses. It fills gaps with assumptions, and they sound convincing.
Too many rights at once
Give an agent write access to every system and errors become hard to undo. Start with reading and drafts.
Every team builds its own agent
Separate agents with their own prompts and sources create separate truths. They need one shared, approved state.
Step by step
- 1
Pick one recurring task
Start with a clearly bounded task whose result a person can check.
- 2
Define the context
Decide which sources and decisions the agent reads, and which it does not.
- 3
Limit the rights
Grant only the tools the task needs. Reading before writing.
- 4
Build in approvals
A person approves anything that leaves the company or changes data.
Bring one important decision.
In 30 minutes with Marc or Simon we show where your work drifts from the decision today.
Meet the Company BrainRead on in the PM Lab
Articles that go deeper.
Related terms
- AI memory →AI memory is the part of an AI system that keeps knowledge beyond a single conversation.
- MCP →MCP (Model Context Protocol) is an open standard through which AI applications access data and tools.
- Human in the Loop →Human in the loop means a person reviews and approves an AI process at defined points before it continues.
- Context Engineering →Context engineering is the work of giving an AI model exactly the right context for a task: the goal, the rules that apply and the knowledge it needs.