Lesson 7: Problem Solving with Vibe Coding and Agentic AI
Risks in using agents
Agents can take actions, not just answer. A mistake can change files, send messages, or publish things.
They can chase the wrong goal. An agent might technically follow instructions but do something unhelpful. (goal misalignment)
They can share private information. If given access, they may use data they should not.
They can waste time or money. Some agents keep trying, looping, or using paid tools.
They can be tricked. A bad website, file, or message might manipulate what the agent does.