Iconoklast/ Independent design & build
Iconoklast.
From an idea to an icon worth refining.
I designed and built a workspace to make generating and iterating on iconography easier—with shared standards, useful alternatives and direct control over the result.
Read the case studyA missing icon. An evolving family.
I started with a simple need: generate and iterate on iconography easily and quickly. I framed two situations that the same workflow needed to support.
- Maya
- A product designer needs a new symbol, with room to explore what it should communicate.
- Alex
- A system designer needs to update a family without repeating every construction decision.
- Shared need
- Reach useful alternatives quickly, then keep control of the detail.

Maya
Product designer“How can I explore this idea quickly?”

Alex
System designer“How can the whole family stay consistent?”
Scenario personas used to frame the design.
Getting an option was only the first step.
The friction continued after the first icon: matching its style, correcting a detail and deciding what to keep. I wanted those steps to feel like one continuous task.
- Starting
- Searching or drawing from scratch delays comparison.
- Repeating
- The same stroke, corner and spacing decisions recur across icons.
- Continuing
- Versions and feedback become harder to follow as alternatives accumulate.
I designed the loop, then built it with Codex.
I took Iconoklast from problem framing and solution ideation into a working web app, vibe coding with Codex and refining the interaction in the implementation.
- Frame
- Make creation and revision quicker without giving up the icon library’s visual language.
- Explore
- Connect a brief, shared standards, alternatives and explicit review.
- Build & refine
- Try the working flow, inspect the generated code and revise the parts that interrupt iteration.
Define the visual language once.
I placed the standards beside the work so each new direction starts from the same constraints. The brief describes the meaning; the standards describe how it should look.
- Construction
- Grid, stroke, corners, spacing and padding shape the family.
- Expression
- Outlined, filled and combined treatments, palette and perspective guide the exploration.
- Reuse
- Presets carry the configuration; written session rules hold the decisions that controls cannot express.
Compare directions. Continue the promising one.
I made alternatives part of the starting flow, then added a way to build on a selected icon. A useful direction should become the basis of the next request.
- Describe
- Write the intent and add visual references by upload, paste or drop.
- Compare
- Explore four to eight directions against the same standards.
- Continue
- Choose an icon, describe what to change or keep, and generate from its geometry.
saved projects.
Redesign the family with its originals in view.
Creating one new symbol and updating a set need different context. For a redesign, I kept each source icon beside its alternatives and shared the direction across the family.
- Preserve
- Keep the original geometry and intended meaning available for comparison.
- Coordinate
- Use common standards, rules and family references across selected icons.
- Review
- Judge both the individual symbol and how it belongs beside its neighbours.
A correction can become guidance. With approval.
I connected review decisions to reusable session rules. A rejection reason or an accepted edit gives the AI evidence to suggest a rule; the designer decides whether that rule is useful.
- Explain
- A written reason makes the requested correction specific.
- Inspect
- Read the proposed rule and change its wording before approving it.
- Apply
- Only approved suggestions join the guidance for future requests. This is session context, not model training.
“Too much detail. Simplify the shape.”
Edit the wording. Any change needs approval again.
Preserve each icon’s meaning.
Suggested rule not included yet.
Written feedback stays with the request.Sometimes the next step is moving a point.
Generating again is not always the right correction. I built direct vector editing into the same workflow so a designer can resolve geometry precisely.
- Edit
- Move anchors and paths, adjust curves, corners and appearance.
- Arrange
- Duplicate, group, align to a reference and distribute repeated geometry.
- Return
- Save a new round for review. Keep the earlier version and confirm that the save succeeded.
Save an edit.
Then review the new round.
Keep what works while exploring what’s next.
Iteration exposed an unnecessary constraint: choosing another direction could undo an earlier choice. I made acceptance independent and kept accepted icons through new rounds.
- Workbench
- Explore alternatives, refine a direction and return to earlier rounds.
- Library
- See the accepted collection, including multiple useful versions of one icon.
- Deliver
- Export editable SVGs with a sprite and manifest. Find saved work again through named sessions and the archive.
The project became about continuity.
The initial idea was to generate and refactor icons. Building it made the space between generations more important: the chosen direction, the correction and the work already accepted.
- Working outcome
- A connected workflow for creating, revising, reviewing and exporting iconography.
- Design learning
- Faster exploration needs precise control and a dependable way to keep progress.
- Next evaluation
- Watch designers complete a real icon task; assess iteration time, correction effort and fit within the family.
An iteration,
with its context.
Explore Iconoklast.
Try the icon workspace, from shared construction rules to editable geometry. The source shows how those decisions are implemented.
Independent project in development. Local features may precede the live app; impact is not yet measured. The case-study illustrations do not send AI requests.