CoachDiscoverMapCompareSavedAI LearningAI WeeklyAbout
← All guides
Guide 02 · Applied

Generating a Blueprint style end to end

Color palette → reference file → new style → applied to a real library. The whole loop, including the "be bold" trick that keeps AI from second-guessing itself.

Walk away with
Use Copilot for the palette
Walk away with
Tell AI to be bold
Walk away with
Random metadata to see it live

Start with a color palette from Copilot

The video kicks off the new style by asking Copilot to generate an image color palette. The output is a blueprint-style image with a swatch strip across the bottom. SharePoint doesn't have to match the image exactly — what matters is the palette swatch and the mood it captures.

Don't write a color palette by hand. Ask Copilot for an image, drop it in your styles library, and reference it.

The image becomes a concrete artifact the next AI call can read. It is much faster to describe "use this palette" than to enumerate hex codes, and it gives the style file a piece of visual evidence to lean on.

"I need a designer to define the palette." For most internal use, a quick Copilot-generated reference image is enough — and it documents intent for the next person who edits the style.

"Be bold" — the prompt that fights AI fence-sitting

AI styling has a known failure mode: the model is trained on every style ever made and starts second-guessing. Did you want subtle? Modern? Avant-garde? When it is unsure, it produces a timid output — restyling just the header or the icons and leaving the rows alone.

Nudge the model with "be bold." It pushes the AI past safe restyling into actual row-level redesign.

The same pattern shows up across AI styling tools in general. Without an explicit nudge, the default behavior is conservative. With one, the model commits — replacing row layouts, changing the whole vibe, taking the kind of swing you actually wanted.

"Vague prompts give the AI room to be creative." They give the AI room to play it safe. Specific, opinionated prompts produce more creative output, not less.

Applying the new style to a library

Once the style file exists, applying it is the same shape as any AI in SharePoint call: a prompt that names the List Styling skill and tells it to use the blueprint style on this library. The author was explicit about which skill to use because their tenant had several style-related skills installed and they didn't want any confusion.

When multiple skills could match, name the skill explicitly in the prompt. Ambiguity wastes a run.

Styling is not the fastest operation — but it is not human-slow either. The model has to read the style file, write the row formatter JSON, apply it to the library, and verify it took. A few minutes per style is normal.

"Once styled, the library is locked into that style." It is not. Apply a different style anytime; the data layer is untouched.

Random metadata to see the style breathe

To demo a new style, the author uploads a few documents and then asks the AI to fill in random metadata. With realistic-looking values across status, owner, due date, and other columns, the style finally has something to render — red overdue chips, blue active rows, the whole blueprint vibe.

A styled library on empty data looks broken. Seed enough realistic metadata to see the rules fire.

This is also a fast way to test conditional styling. If "overdue" should render red, you need an overdue item. If you want to see the green status pill, you need an item in the right state. AI-generated random metadata is the cheapest way to get there.

"I need real production data to evaluate a style." You don't — synthetic data is enough to validate that the style fires the right rules and reads well at a glance.
Flashcards — Applied
02 · Applied
Color Palette Image
tap to reveal →
A Copilot-generated reference image that establishes the color palette for a new style. Dropped into the styles library and referenced by the next prompt.
← tap to flip back
02 · Applied
"Be Bold" Prompt
tap to reveal →
An explicit nudge in the prompt that pushes the AI past safe restyling. Without it, the model fence-sits and only restyles headers or icons.
← tap to flip back
02 · Applied
Blueprint Style
tap to reveal →
The monochromatic-blue, technical-drawing style generated in the demo. Built by combining a palette image, a reference style file, and a "be bold" prompt.
← tap to flip back
02 · Applied
Explicit Skill Naming
tap to reveal →
Calling out which skill should run when multiple skills could match the prompt. Prevents the agent from picking the wrong styling skill.
← tap to flip back
02 · Applied
Random Metadata Fill
tap to reveal →
Asking the AI to populate test rows with realistic synthetic values. Lets a new style render against meaningful data without real production records.
← tap to flip back
Guide 02 · Applied · Building a Blueprint Style All flashcards →