AI Agents
An agent works toward a goal instead of waiting for every next prompt.
You give an outcome. The system can decide intermediate steps, use tools, inspect results and continue until it reaches a stopping point.
Chatbot
You steer each step
1
Search for vendors.
2
Now find pricing.
3
Now compare them.
4
Now draft the email.
Agent
You give the outcome
“Research 3 office-furniture vendors and email me a comparison by 5pm.”
The system can plan and execute the intermediate work within the permissions and boundaries it has.
The agent loop actually loops
1 · Observe
Read goal and current state.
2 · Plan
Choose the next move.
3 · Act
Use a tool or create output.
4 · Check
Is the goal complete?
↩ If the goal is not complete, the agent plans another step.
Ready.
Watch the agent handle an obstacle
Key insight: autonomy comes from repeated observe → plan → act → check cycles. The agent can make bounded intermediate decisions without asking for every tiny step.
AI Agents · Go deeper
An agent needs a goal, intelligence, tools and working state.
These ingredients determine what the agent can decide, what it can do and how it keeps track of progress.
Four ingredients
Tap an ingredient
See what role it plays.
Autonomy is a spectrum
Your role
AI role