Table of Contents
08. How to Ask for Safe Changes
How we request improvements without triggering refactors, redesigns, or unintended rewrites
—Part of CSS Primer for the AI Era — CSS Communication Toolkit
This chapter is the practical heart of the Toolkit. It teaches us how to request changes in a way that keeps AI from over‑reaching, over‑interpreting, or rewriting the system.
AI does not understand the difference between:
- a small adjustment
- a refinement
- a correction
- a fix
- a redesign
- a refactor
Unless we name the scale of the change,
AI assumes it has permission to “improve” the system.
So, the eighth communication skill is simple:
We ask for changes in a way
that keeps the work small,
predictable, and safe.
Safe changes are not about limiting creativity.
They are about protecting the system while still moving forward.
1. Why Safe Changes Require Explicit Framing
When we say:
“Fix this”
“Improve this”
“Make this responsive”
“Clean this up”
AI must decide:
- how big the change should be
- how deep the change should go
- how many related patterns to unify
- whether to refactor the layout
- whether to reorganize the CSS
- whether to rewrite the HTML
A simple truth:
AI expands the change
unless we define the scale.
So we define it.
Let's see how.
2. The Three Scales of Change We Need to Name
Every request falls into one of three scales:
- Adjustment
A small, local change. - Refinement
A slightly broader improvement within a defined boundary. - Revision
A larger change that still respects constraints.
We name the scale
before we describe the intention.
3. How We Ask for an Adjustment (Smallest, Safest)
An adjustment is a tiny, precise change.
We might say:
“We want a small adjustment.”
“Make a minimal change.”
“Keep the solution as small as possible.”
“Do not modify anything outside this rule.”
Examples:
“Adjust the spacing by one step in the scale.”
“Center this element without changing the HTML.”
“Reduce the shadow slightly.”
Adjustments keep the system stable.
4. How We Ask for a Refinement (Still Safe, Slightly Broader)
A refinement improves something
while staying inside a boundary.
We say:
“We want a small refinement.”
“Improve this without changing the structure.”
“Stay within this component.”
“Do not introduce new patterns.”
Examples:
“Refine the alignment within this group.”
“Improve the rhythm using the existing spacing scale.”
“Make the card feel quieter without changing the layout.”
Refinements keep the system coherent.
5. How We Ask for a Revision (Larger, But Still Controlled)
A revision is a larger change,
but still bounded.
We say:
“We want a controlled revision.”
“You may adjust multiple rules, but stay within this component.”
“Do not change the HTML or global patterns.”
“Keep the behavior stable.”
Examples:
“Revise this layout to use grid, but keep the structure intact.”
“Revise the spacing pattern while staying within the scale.”
“Revise the visual tone to feel quieter, but do not change hierarchy.”
Revisions move the system forward
without breaking history.
6. How We Prevent AI From Over‑Reaching
We add explicit safety language:
“Do not refactor.”
“Do not redesign.”
“Do not unify patterns.”
“Do not reorganize the CSS.”
“Do not introduce new classes.”
“Do not modify unrelated rules.”
A simple principle:
Safe changes are safe because
we name what is not allowed.
7. How We Anchor the Change With Boundaries
We combine:
- the scale
- the target
- the context
- the exclusions
For example:
“We want a small adjustment. Center this element horizontally. The HTML is context — do not modify it. Do not change other spacing rules.”
Or:
“We want a refinement. Improve the rhythm within this group. Stay inside this component. Use the existing spacing scale.”
Or:
“We want a controlled revision. Convert this layout to grid. Keep the structure and behavior unchanged. Do not modify other components.”
This is how we keep the work safe.
8. How Safe Requests Reduce AI’s Cognitive Load
When we ask for safe changes:
- AI stops scanning the entire file
- AI stops trying to unify patterns
- AI stops refactoring
- AI stops redesigning
- AI focuses on the target
- the change stays small
- the system stays stable
A simple reminder:
Safety is not caution.
Safety is clarity.
9. Closing Note — Safe Changes Are the Foundation of Trust
When we ask for safe changes clearly:
- we protect the system
- we protect the design
- we protect the history
- we protect the collaboration
- and we protect our own agency
A simple closing thought:
We define the scale of the change.
AI works within it.
And the system remains whole.
- Suggested Next Reading: How to Ask AI to Explain Its Choices
Tony de Araujo —New York | Lisbon