Daily AI CatchupAgent or Not?
Five scenarios. Decide what actually makes each one "agentic" — or not.
About 2 min to play
0 / 5
What actually makes something an "agent" rather than a single model call?
An agent uses a browser tool to check a flight price, then books it. What just happened at the "check the price" step?
What's the safer default for an agent that can send real emails on your behalf?
What typically stops an agent loop from running forever?
A tool call an agent makes fails partway through. What should typically happen?
Why this matters
"Agent" gets used loosely enough that it's worth pinning down. A single call to a model — you send text, it sends text back — isn't an agent, no matter how good the answer is. What makes something agentic is a loop: the model takes an action, sees what actually happened as a result, and decides what to do next based on that — not on a script you wrote in advance.
That loop is also where the real design decisions live. How much should it be allowed to do without checking in? What happens when a tool call fails — does it retry, try something else, or stop and ask? What stops it from running forever if it never quite finishes the task?
None of this is exotic engineering. It's closer to normal software design — error handling, permission boundaries, exit conditions — applied to a system whose next move you can't fully predict in advance.
What you'll learn
- The defining feature of an agent is the loop — act, observe the result, decide the next action — not model size or prompt phrasing
- Tool use means the model triggers a real action and reads back a real result, not just describes one
- Consequential actions (sending, spending, deleting) deserve a human checkpoint by default, not just logging after the fact
- A real exit condition and a step limit are what stop an agent loop from running forever
This game goes with What Actually Makes Something an "AI Agent".