What not to ask it yet, and where to go next

You've seen what an agent does well. So you don't waste an afternoon, it's also worth knowing where the limit is today, especially with free models.

What it does well

Reading and summarizing a folder, a document, or a data file.
Small, well-described changes: creating a file, fixing a format, renaming something.
Explaining what a file or a command you don't understand does.
Proposing a plan before you decide.

Huge, vague jobs, like "build me a complete app". It gets lost, and what it delivers is hard to review. Break the work into small steps.
Things you can't check. If you have no way of knowing whether the result is right, don't accept it blindly.
Touching important folders (your work, your personal documents) while you're learning. Practice in practice folders.
Personal or confidential data with free models: several use what you write to them for training.
Commands that delete or install things without you reading first what they'll do. That's why we left "bash": "ask" in the configuration.

Practice this week with the exercises in the After module. Ten minutes a day is worth more than one long afternoon.
The October workshops, every Wednesday at 5:30 p.m. Mexico City time: Your first AI API: connect an LLM with Python; From chat to agent: make an AI use tools; Connect your agent to real tools with MCP; and Your first model running on your laptop. They're in the Workshops section of the platform.
The AI Engineering Bootcamp, if you want to go from using agents to building them. The information is in the Bootcamps section.