Research Demos

TikTok for PR Reviews

Seven interactive prototypes exploring how to make code review faster and more tractable — from short-form content UX patterns to AI-driven context layers that surface why decisions were made, not just what changed.

Research Context
Why these demos exist
The problem, the two-surfaces framing, notes on each demo, and key insights from the research. Start here if you're seeing this for the first time.
Background Key insights
Read →
Demo 1
Swipe Card Feed
Each PR is a card. Swipe or press arrow keys to approve, request changes, or snooze. Large PRs require an extra confirmation step.
Touch / keyboard PR triage Friction guard
Open →
Demo 2
Diff Reel
File-by-file vertical feed with snap scrolling. Per-file decisions, auto-advance countdown, and a reviewability score for each PR.
Scroll snap Auto-advance Per-file review
Open →
Demo 3
PR Size Coach
Author-side tool. Paste a file list and get a Reviewability Score with split suggestions, streak tracking, and a team leaderboard.
Author UX Score gauge Gamification
Open →
Demo 4
"For You" Feed
Algorithm-sorted review queue. Toggle algorithm ON/OFF to see the difference vs. FIFO — the 1,800-line PR moves to the top instantly.
Algorithm Queue sorting Explainability
Open →
Demo 5
The AI Brief
For AI-generated PRs: the agent briefs the reviewer in first-person — decisions made, alternatives considered, and what it flagged for human judgment. Toggle agent-generated vs. AI-inferred, and see the same 347-line PR with and without the brief.
AI agent Decision surface Context layer
Open →
Demo 6
The Decision Surface
For human-written PRs with no author context: AI infers likely decisions from diff shape and generates skeptical questions a thorough reviewer should ask. Reviewer marks each question as answered or flags it for the author.
Human PR AI inference Structured review
Open →
Demo 7
The Agent Story Arc
What becomes possible when an AI agent writes rich commit messages by default. Open with the same PR shown as human commits ("fix auth", "wip") vs. agent commits with decisions, tradeoffs, and risk at each step — then review the agent version step-by-step.
Agent norm Commit story Human vs. agent
Open →