Case study · No-code editor

Candu is a live, shipping no-code editor for building in-product tours and banners. I founded its design system from scratch, left for another role, then came back to lead its move into AI-assisted creation, most visibly with Recording Session: a Chrome extension that turns a manual, step-by-step tour build into a single recorded flow the AI drafts and the user approves.

Role Founding & returning solo product designer

Team 2 engineers, no dedicated PM or QA (same team building Dubot)

Timeline 2019–2022, 2023–present

Status Live in production

The problem

Candu's users are non-technical marketers and product teams building in-product tours and banners without writing code. Before Recording Session, creating a tour meant building each step individually: connecting it to a placement, defining the target URL, writing the content, and styling every step's look and feel by hand. A simple 5-step tour took 30 to 60 minutes, done by people who are not designers and often just want to ship a quick walkthrough.

Research here didn't come from a large quantitative study. Candu is a B2B power-user product with a small number of people creating content for many end users, so the signal came from user interviews, support tickets, CX feedback, and in-product usage patterns, like which editor components got used most, which also fed a separate palette-restructuring project. That's a different research mode from a consumer product's funnel numbers, and just as valid.

The solution

Recording Session is a Chrome extension that captures a real flow as the user performs it once in their own product. The AI-powered capture identifies each meaningful interaction as it happens, with visual feedback (a countdown, step indicators, a halo effect on captured elements), then generates the tour steps automatically: titles, copy, selectors, and screenshots pulled straight from the recording.

The designed recording-complete state, ready to hand off to the editor.

The user lands in the editor with a real draft already built: steps positioned on the actual product, copy already written, ready to fine-tune.

The designed editor result: an AI-drafted tour, ready for human review before publishing.

The interactive walkthrough below simulates that same flow end to end, so you can click through it yourself.

Interactive walkthrough Step through the toolbar above the demo: idle → record → recorded → AI draft → publish

This is a simplified simulation built to demonstrate how the flow works, not a pixel-accurate recreation of Candu's UI. The two screens above are the real designs it's based on.

Key decision

AI drafts the tour, but nothing publishes automatically. The user reviews the AI-generated copy and structure on the real product and approves before anything goes live. It's a deliberate quality gate, so the speed gain doesn't come at the cost of quality.

Outcome

Recording Session is live in production today. Power-user feedback has been directionally positive, though at Candu's size there's no dedicated team tracking that at the level of individual tickets or quotes, a real constraint of running a small, lean team. Alongside this flagship flow, I also led the migration of Candu's design system to a tokenized framework, standardizing UI across the internal tools and the newer AI agent surfaces.

I design this work, and run Candu's broader design documentation and process, working with Claude Code as a collaborator. The human-in-the-loop pattern in Recording Session mirrors how I actually work day to day.