Fearless iteration: one change at a time, roll back freely

A beginner looks at the first version and writes: "This is all wrong, redo it." An hour later he's on his fifth "all wrong". A pro moves differently.

Rule 1: one change at a time

"Move the form higher" → look → "now make the headline bigger" → look. Changing one thing at a time, you see the effect of each step and never lose the bits that worked. A bundle of ten edits is a lottery: something improves, something breaks, and you can't tell which is which.

Rule 2: specifics, not verdicts

• ❌ "It's ugly" → ✅ "Too cramped: add breathing room between the blocks"
• ❌ "It's boring" → ✅ "Add one bright detail to the header — say, an accent bar with the offer"
• ❌ "It doesn't work" → ✅ "I click Send — the form doesn't clear and there's no confirmation"

Rule 3: rollback is your superpower

Lovable and Replit keep a version history: any edit can be undone with one button. That changes the psychology: you can't break anything. Experiment freely — the worst outcome is stepping back one move.

When the AI goes in circles

You ask for a fix — something else breaks. You fix that — the first thing breaks again. Meet the death loop. The way out: roll back to the last working version and re-state the task from scratch, whole and differently: "Rebuild the schedule block from zero: a table on desktop, cards on mobile." A fresh statement beats a tenth patch.

Why the first version is always a draft

Even seasoned designers don't draw the final on the first pass. At Apple, by former employees' accounts, every screen went through dozens of iterations before it shipped. The difference between a beginner and a pro isn't that the pro "gets it right immediately" — it's that the pro isn't upset by a draft and knows how to walk it toward the goal in small steps. You're in exactly that position, except your iterations take seconds instead of days, because the machine does the editing.

A classic mistake: the ten-edit bundle

The most common beginner trap looks like this: write down everything you dislike and send it as one ten-point message. The AI does it — and some things improve while others break, with no way to tell which point caused what. So the person sends ten more edits on top, and now it's hopeless. The fix isn't "phrase the bundle better", it's don't build a bundle at all: one edit → look at the result → next. Slower on paper, three times faster in practice, because you never roll back into chaos and always know what changed.

Common myths

• "If the version is bad, wipe it and start over." Almost never. A bad version is usually 80% keepers and 20% problems — roll back surgically instead of torching the project.
• "Lots of edits means I described it badly." No: lots of edits is how the process works, not a sign of failure. What's bad is edits that are random and leave you unsure what improved.
• "The AI will get annoyed or tired if I keep asking." It won't. Ask as many times as you need — one clear edit at a time.

Templates for concrete edits

Keep phrasings the AI can't misread within reach:

"Double the breathing room between the pricing and testimonials sections."
"Make the Book now button bigger and move it to the first screen."
"After the form is sent, show the message 'Thanks, we'll call you'."
"Rebuild the schedule block from scratch: cards on phones, not a table."

A Socratic pause

Do this now

Take any site you use every day and write three edits for it using "specifics, not verdicts": not "it's awkward", but exactly what's cramped, tiny or unclear. That's the language you'll speak to the AI in the next module, when you open Lovable and build your first landing page. Notice how hard it is at first to turn "I don't like it" into a checkable fact — that's the skill separating the people for whom "AI just works" from the ones who say "the tool is dumb".