From 22 buttons to one decision

The AI panel inside the Alex CMS editor used to show every action it could take as an equal button: around twenty in the editor, more behind "show all". All of them worked, and I still found it confusing, and I built it. This is what changed and why.

What was wrong

• No hierarchy. "Generate image" and "Fact-check content" looked equally important on a page that had never been fact-checked.
• Labels described the tool, not the result: "Complete metadata", "Generate final SEO". You had to know the system to know what you'd get.
• The scores sat at the top as dashes, with nothing connecting an empty score to the button that fills it.
• The panel never answered the one question an editor actually has: what should I do next?

What changed

• Page health leads, in plain words, and each unscored metric is a one-click check.
• One recommended next step, chosen from the page's actual state, with the reason written next to it.
• Everything else is folded into four stages of finishing a page: Create, Improve, Verify, Publish. The stage you're in opens by itself.
• Every label was rewritten outcome-first: "Fact-check it", "Check it sounds like you", "Fill in tags & slug". Routing still keys off the stable action IDs, so no behaviour changed.
• Typing what you want still works for people who'd rather not click at all.

What I'd still test

I'm the only editor, so this is judgement, not a usability study. The honest next step is watching someone else finish a page with it, and timing how long it takes from an empty draft to published.