Agentic Coding Workflows

Lesson 1 of 6

What Makes an Agent Different from Autocomplete?

A prompted assistant answers one question and stops. An agent keeps going: it reads files, runs commands, looks at the results, and decides what to do next, on its own, in a loop, until the task is done.

The agent loop

Most agentic coding tools (Claude Code, Cursor's agent mode, Copilot Workspace) run something like this:

1. Read the task and relevant files.
2. Decide on an action (edit a file, run a command, search the codebase).
3. Take that action using a tool.
4. Observe the result (did the test pass? did the command error?).
5. Repeat, until the task looks done or it needs your input.

That "observe and repeat" step is the key difference. A one-shot prompt can't notice that the test it just wrote is failing and go fix it. An agent can.

Tools are what make this possible

An agent isn't smarter than a chat model, it just has tools: functions it can call to read a file, write a file, run a shell command, or search the web. Every action you've seen an assistant take, editing three files, running your test suite, fixing a failure it just caused, is a tool call, not magic.

Chat / prompted generationAgentic loop
StepsOne request, one responseMany steps, chosen by the agent
Sees results?No, it just produces textYes, reads output and reacts
Can fix its own mistakes?Only if you point them outOften, before you even see them
Your roleWrite the prompt, read the answerSet the goal, review the plan, approve risky steps

Why this changes how you work

Because an agent can take many actions unsupervised, your job shifts earlier in the process: from reviewing a finished answer, to scoping the task well and putting guardrails on what it's allowed to do, before it starts. The rest of this course is about that shift.

NOTE

None of this makes the agent less accountable to you. It just means the checkpoints move: instead of only reviewing the output, you'll also review the plan, the tools it's allowed to use, and the diff, at different points along the way.

📝 Agent Basics Quiz

Passing score: 70%
  1. 1.What is the key difference between a one-shot prompted response and an agentic loop?

  2. 2.What lets an agent take actions like editing files or running commands?

  3. 3.The agent loop is: read, decide, act, ____, and repeat.

  4. 4.An agent can notice that a test it just wrote is failing and attempt to fix it without you re-prompting it.

  5. 5.Because agents can act unsupervised, your role shifts partly to scoping tasks and setting guardrails before the agent starts.