Lesson 4 • Agents

Agent capability rises fast.
So does risk.

Toggle permissions like email, files, browser and money, then watch how an AI agent becomes more useful, more autonomous, and more dangerous.

Interactive Simulator

Agent Risk Simulator

Choose a scenario. Then add permissions. The simulator updates the agent’s capability, autonomy level, and risk profile in real time.

Step 1 • Pick a scenario
Step 2 • Toggle permissions
Live Readout

Current agent profile

Balanced
Capability 0 / 100
Risk 0 / 100
Autonomy
Manual assistant
Primary failure mode
Low leverage
Recommended guardrail
Human review
Agent summary
Careful research helper

This agent can help gather and organise information, but it still relies heavily on a human to decide, approve, and act.

Permission impact

What is making this risky?

Key Lesson

Useful agents are not the danger. Unchecked agents are.

Adding more permissions can turn a chatbot into a real operator. That is where approvals, budgets, scope limits, audit trails and sandboxing start to matter.

Least privilege

Give the agent only the minimum permissions needed for the task in front of it.

Human checkpoints

Keep approvals on anything involving money, publishing, deletion, external messages or legal exposure.

Logs and limits

Track every action, cap spend, define safe tools, and make it easy to stop the agent fast.