Everything so far has been AI that responds when you talk to it, then stops and waits. The newest shift is AI that keeps working after that single message — checking its own progress, using tools, and deciding what to do next without you typing every single step. That’s an agent.
Chatbot vs. agent, in one sentence
A chatbot: you ask, it answers, done — one turn, then it’s your move again. An agent: you give it a goal, and it keeps taking steps — checking results, adjusting, trying again — until the goal is actually finished, not just discussed.
- 🎯 Get a goal
- 🤔 Plan a step
- 🛠️ Use a tool
- 👀 Check result
…and loop back to “plan a step” until the goal is actually done — no human needed to re-prompt each stage.
The other half: using real tools
An LLM on its own can only produce text. What makes an agent actually useful is pairing it with tools it can call on: searching the web, running code, reading a file, sending an email, booking something. The model decides when a tool is needed and which one — then reads the result before deciding its next move.
That’s the real shift from part 1’s “guess the next word” — an agent’s next move might not be a sentence at all. It might be “open this file” or “run this search,” with the actual writing saved for the very end.
where it shines, where it stumbles
Where it shines
Multi-step chores with a clear, checkable finish line — research, scheduling, data cleanup, running tests.
"Find flights under $400 and shortlist three" is a real agent task.
Where it stumbles
Ambiguous goals, high-stakes actions (payments, deletions), or tasks with no clear "done" — it can wander or loop.
This is why most agents still ask before anything irreversible.
The common thread: agents earn trust the same way a new hire does — small, checkable tasks first, with a human glancing at the results before the leash gets any longer.
where you've met it already
- Coding agents — Writing, running, and fixing code across a whole project
- Research assistants — Running dozens of searches, then compiling one clean report
- Inbox & calendar bots — Triaging email, finding a meeting time across everyone’s schedule
- Customer service — Looking up your order, issuing a refund, closing the ticket — no human handoff