Lesson 7 of 8
Beyond the chat box
6 min read
A normal chat answers in seconds. But some jobs are ten minutes of planning, doing, and checking. What if you could hand Claude the goal and let it work?
Claude that acts in a loop
Beyond the chat box, Claude can act — not just reply. Give it a goal and it works in a loop: think about the next move, act with a tool, observe what came back, then think again. It keeps going, adjusting when something doesn't work, until the goal is met.
An agent doesn't need every step spelled out. Give it the goal; it works out the steps and adapts when one fails.
The same loop, everywhere
That loop shows up in a few places. Claude Code runs it in your codebase — planning and editing files. Claude in Chrome runs it in your browser — clicking and filling things in. Cowork runs longer jobs alongside you. Different surfaces, the same think-act-observe engine.
Acting is a trade of speed for reach: minutes of autonomous work in exchange for a result you'd have spent an afternoon on.
An agent acting on its own can also be wrong on its own. Keep it on a short leash for anything that spends money, sends a message, or can't be undone — read what it did before you approve it.
The shape of it
- —Claude can act in a loop — think, act, observe — until a goal is met.
- —Claude Code, Claude in Chrome, and Cowork run that same loop in different places.
- —Great for multi-step jobs — but check its work on anything risky.
You ask Claude to "book a table for Friday night" and let it work. What does it actually do?
Continue in the app
Take the whole Get the most out of Claude course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.