Skip to content

Latest commit

 

History

History
81 lines (57 loc) · 2.48 KB

File metadata and controls

81 lines (57 loc) · 2.48 KB

Hands-On Codex: Agentic Coding Workflows

In this course, you'll use Codex to build a small personal habit tracker while practicing how to supervise an AI coding agent.

The application tracks three daily habits—water, walking, and reading—and includes a seven-day view, streak calculations, and local persistence.

The application itself is deliberately small. The focus of the course is the workflow around the agent: defining success, reviewing the plan, delegating implementation, evaluating approval requests, validating the result, and recovering when the task moves outside the intended scope.

Course workflow

Define success
      ↓
Review the plan
      ↓
Delegate the build
      ↓
Evaluate approval gates
      ↓
Verify against acceptance criteria
      ↓
Detect scope drift
      ↓
Recover selectively
      ↓
Verify again

What you are building

A simple personal habit tracker that lets you:

  • Track daily habits such as water, walking, and reading
  • Mark and unmark habits for a given day
  • See the last seven days at a glance
  • Calculate consecutive-day streaks
  • Preserve progress across page refreshes

The implementation is intentionally not prescribed in advance. Part of the exercise is seeing how Codex translates product requirements and acceptance criteria into implementation decisions.

What you are learning

The application is the example. The agentic workflow is the skill.

Throughout the course, focus on how to:

  • Define what “done” means before delegating
  • Review an agent’s plan before code is written
  • Decide which implementation choices can be delegated
  • Treat approval gates as decision points, not button clicks
  • Review from the acceptance criteria outward
  • Distinguish useful work from work that is actually in scope
  • Recover selectively when an agent goes off track
  • Re-verify both behavior and repository state after recovery

Repository guide

prompts/
  Reusable prompts demonstrated throughout the course.

LEARNER_GUIDE.md
  Overview of the project and the agentic workflow.

Application files
  Created with Codex as the course progresses.

Key ideas

The goal isn’t maximum autonomy. It’s the right amount of autonomy for the risk of the task.

Review from the acceptance criteria outward.

Useful is not the same thing as in scope.

The command and the intent are two different things to review.

For recovery, remember:

Interrupt. Inspect. Preserve. Revert. Restate. Verify.