← Back to work

Case study · No-code editor

Cutting a 30–60 minute manual workflow down to under 10.

Candu is a no-code editor for building in-product tours and banners. I was its founding designer, and I came back in 2023 to lead its move into AI-assisted creation.

  • Founding designer, now solo product designer
  • 2 engineers, no PM or QA
  • 2019–2022, 2023–present
  • Live
  • candu.ai ↗
The challenge

Building one tour took 30–60 minutes of manual setup, done by people who aren't designers.

What I did

Designed Recording Session: record a flow once, AI drafts the tour, the user reviews it before it goes live. Also moved the design system to tokens.

The result

Live in production. A tour now takes under 10 minutes.

The problem

Building one tour meant 30–60 minutes of manual setup, every time.

Every step had to be linked to a placement, given a URL, written and styled by hand. I learned this from interviews, support tickets, CX feedback and usage data.

The solution

Record the flow once. Let AI draft the tour. Review before it goes live.

A Chrome extension captures the flow as the user clicks through it once, then generates the steps: titles, copy, selectors and screenshots.

Designed recording-complete state of the Candu extension: four captured steps filled in, a Tour Flow Recorded confirmation, and a Save and edit in Candu button.
Recording complete, ready to open in the editor.
Designed Candu editor showing an AI-drafted tour step on the target product, the AI Copilot's summary panel, and a floating step navigator.
The editor opens with a real draft on the product, ready to review.
Interactive walkthrough Use the toolbar to step through the flow

A simplified simulation of the flow. The two screens above are the real designs.

Key decision

The AI could have published tours on its own. I added a review step instead: the user checks the draft on the real product and approves it first. 10x faster, not 10x worse.

Hiring for a product design role? I'd love to hear about it.