The Vibe Coding Workflow
A day in the life of an AI coder. The simple loop you'll repeat to add features, fix bugs, and ship without breaking things.
In this chapter
- The daily loop every AI coder follows
- How to add a new feature without breaking the old ones
- How to fix a bug without making it worse
- When to save, when to undo, and when to start fresh
Real World Example: Adding a Newsletter Signup
You want to add a newsletter signup box to the bottom of your homepage. Instead of asking AI to do it all at once, you break it into three small prompts: build the box, wire up the form, then style it. After each step you check the browser and save with Git.
“On my homepage, add a newsletter signup section just above the footer. It should have a heading that says 'Stay in the loop', a short subtitle, an email input, and a 'Subscribe' button next to it. Match the purple gradient style I'm already using, and make it stack vertically on mobile. Don't connect it to anything yet, just the visual layout for now.”
The Daily Loop
Every AI coder follows the same five-step loop. Master this and everything else gets easier.
A Day in the Life
Here's what a typical building session looks like, start to finish.
How to Add a New Feature
Big features are just lots of small ones. Break them down before you prompt.
-
1
Describe the feature in one sentence Before opening AI, write down: 'Users should be able to bookmark articles and see them on a Saved page.' If you can't say it in one sentence, it's not ready yet.
-
2
Break it into tiny pieces List the steps: (1) bookmark icon on each article, (2) remember it when clicked, (3) show the saved ones on a Saved page, (4) show a friendly message when nothing is saved. That's 4 prompts, not 1.
-
3
Prompt the first piece only On the article cards, add a bookmark icon button in the top-right corner. When I hover over it, it should turn purple. Don't make it actually save anything yet, just the icon and the hover effect.
-
4
Check it in the browser Open your project in the browser, scroll through the article list, and hover over the new icons. Then click around the rest of the page to make sure nothing else broke.
-
5
Save with Git Commit with a clear message: 'Add bookmark button to article cards'. Now the next prompt has a safe starting point.
-
6
Move to the next piece Now ask: 'When someone clicks the bookmark button, remember which article they saved so it shows up later, even if they close the tab.' Then check it works and save again. Keep going until the whole feature is done.
How to Fix a Bug
Bug fixes follow a different rhythm than features. Reproduce first, then prompt.
Don't do this
- "The page is broken, fix it."
- Pasting your whole project at AI
- Asking AI to guess what's wrong
- Letting AI "fix" without seeing the error
- Skipping the browser check after the fix
Do this instead
- Reproduce the bug yourself in the browser
- Copy the exact error message word for word
- Tell AI which page it's on and what you expected
- Ask for the smallest possible fix
- Test the fix AND test the rest still works
The bug-fix prompt template
"On the [which page or section], the [which thing] is doing [what's wrong]. It should be doing [what you expected]. Here's the error message I'm seeing: [paste error]. Please make the smallest change you can to fix it."
Routine Tasks Every AI Coder Does
Beyond features and bugs, here's the daily housekeeping that keeps a project healthy.
When the Loop Breaks Down
Some sessions feel cursed. Here's how to recover instead of rage-quitting.
Copy-Paste Prompts
Real prompts you'll use every day. Swap the words in brackets for your own details. Describe things the way you'd describe them to a friend.
“On my [which page], add a [what you want, e.g. 'small contact button in the top right corner']. Match the style of the rest of the page. Please don't change anything else, and tell me in plain English what you did.”
“On my [which page], the [which thing, e.g. 'sign-up form'] is doing [what's wrong]. It should be doing [what you expected]. Here's the error message I'm seeing: [paste error]. Please make the smallest change you can to fix it.”
“On my [which page], please give the [which section] a bit more breathing room, make the heading bigger, and add a subtle effect when I hover over the cards. Keep the same colors and the same overall layout.”
“My project is getting a bit messy. Without changing how anything looks or behaves, can you reorganize things so they're cleaner and easier to work with? Tell me in plain English what you changed.”
“Before I keep going, can you explain in plain English what you just changed and why? Keep it short: three or four bullet points, no jargon.”
“That last change broke [what broke]. Please undo it. Then try a different way to [what you originally wanted] that doesn't touch [the thing that broke].”
“I'm wrapping up for the day. Please save all my changes with a short message that describes what I worked on today. Keep it simple: one clear sentence.”
“Look at where my project is right now and suggest the next three small things I could work on tomorrow. For each one, give me a short description in plain English. Put the easiest one first.”
Frequently Asked Questions
What does a typical day of AI coding look like?
You open your project, decide on one small thing to add or fix, write a clear prompt, let AI generate the change, check it in the browser, test that nothing else broke, then save your work with Git. You repeat that loop, one tiny change at a time, until you're happy with the result.
How small should each prompt be?
Small enough that you can describe it in one or two sentences and check it in under a minute. 'Add a contact form' is too big. 'Add a name input field to the contact form' is just right. Smaller prompts mean fewer surprises and easier fixes when something goes wrong.
When should I save my work with Git?
After every change that works. Think of Git commits like save points in a video game. Every time you reach a working state, save. That way if the next prompt breaks something, you can always go back to your last save point instead of losing hours of work.
What if AI breaks something that was already working?
Stop. Don't ask AI to fix the fix; that often makes it worse. Instead, undo to your last working save point (with Git or your editor's undo), and try a different prompt. The cleaner the starting point, the better the next attempt.