MIKE OICHERMAN
PRODUCT DESIGN · RESEARCH · AI-ASSISTED PROTOTYPING & BUILD
REPLAY.IO BUILDER
Back
An AI vibecoding tool that plans, designs, and builds fully working web apps and fixes its own bugs.
Role Product Designer Consultant
Timeline Q4 2025 - Q2 2026
Company Replay.io
The Problem
Most AI coding tools stall the moment something breaks: they don't see what actually happened at runtime, so anyone using them gets stuck debugging code they didn't write. Replay Builder is built around Replay's deterministic browser-debugging engine, so it can trace exactly what happened inside an app and repair itself.
Goals
I was hired as a consultant to research UX patterns and user behavior on a pre built prototype version created by the team at Replay.io, following by creating a design system and a UX/UI rework.
OBJECTIVE 1 Identify product market fit & analyze user behavior and usage patterns.
OBJECTIVE 2 Create a design system & design 1st iteration based on the research.
Research & Findings
Goals To map the creator's journey, identify user needs, pinpoint high-friction states, and uncover the "AHA" moment.
Process
70x Live User Interviews Conducted with highly engaged users to understand their current pain points and usage habits with back then emerging GenAI technology.
500+ Screenshare recordings reviewed Observed real user recordings to identify unlearn UX patterns, frustration points & usage patterns.
Competitive Analysis Investigated how competitors differentiate themselves and identified potential feature gaps in our offering.
Key Findings
An ‘Empty Canvas’ can be terrifying, as most of the vibe coding tools the user was met with an open ended start point → “What would you like to build?”; While some users found this liberating and explored ideas, the other side of the spectrum didn’t know where to begin. The forked experience between LLM chat vs ‘classic’ canvas behavior created a dissonance in some users when interacting with the app for the first time. Most of the users had a clear expectation of what Builder should do, and have had experience with vibe coding tool, indicating somewhat of an oversaturation in the market. LLM input can take you so far, users wanted to control specific pages, sections and components without having to burn through tokens.
Prototype
Planning While we have continued performing interviews on the live application, we have started exploring the next iteration, focusing on greater creative flexibility, control and sparking imaginations.
Prototype Creation I developed a detailed high-fidelity prototype to simulate the core experience. This was showcased to 50+ unaffiliated interviewees.
Providing a library of templates and examples Reduce anxiety by providing inspiration for first time users & demonstrate the capabilities of builder
Customization & Control Allow users to tweak visual elements without needing to re-prompt the whole project.
Testing & Validation
Method 50x 60-minute video interviews with screen-sharing, where users interacted with the prototype, sharing their thoughts and feedback
Key Findings & Iterations
Templates are powerful, Maybe a little bit too powerful. Some user’s just went to straight up to duplicate an apps and showed little interest in customizing the end product.
Trying catering everyone, serves nobody. In real world simulations only a small percentage of the interviewees wanted to interact with canvas objects and design elements, most of the stack used LLM input as main POC.
Design System
Foundation I created and maintain the design system, built on the Shadcn framework and mapped to brand guidelines, giving the product a consistent visual language grounded in a well-supported, extensible component base.
Velocity Working with AI-ready components let design and engineering move quicker, components were built to be legible to both designers and AI coding tools, shortening the gap between design and shipped UI.
Coverage Beyond the base framework, I created a set of additional components to support UX patterns unique to Loop QA, covering interactions the underlying framework didn't account for out of the box.
Live Product View Product