Glossary
What is review bottleneck?
The review bottleneck is what's left after an agent loop multiplies output: humans can generate a PR, a preview, or a build in seconds, but someone still has to look at it before it ships. Speeding up generation doesn't remove that constraint: it moves the constraint from writing to checking, and checking doesn't scale the same way.
Why it matters
A person can only review so many screens, diffs, or staging links per day, no matter how fast the agent behind them runs. When a loop 10x's its output and review capacity stays flat, one of three things happens: the queue backs up, someone starts rubber-stamping to keep pace, or unreviewed changes ship anyway. All three show up as the same symptom teams call "AI slop": not bad code, unchecked code. Naming the bottleneck matters because it tells you where to invest: not in a faster loop, but in a cheaper review step.
How Patchrooms uses it
Patchrooms doesn't try to remove the review step: it compresses it into one click. A reviewer opens the live preview, clicks the broken element, types or records a comment, done; no account, no ticket to file. "Copy agent feedback" turns that click into a Markdown prompt block (selector, goal, constraints, screenshot ref) the next loop iteration can read and act on directly. The MCP server (list_reports, get_report, set_status) lets the agent pull its own review queue instead of waiting on someone to relay it, so the bottleneck shrinks to the one thing that actually needs a human: judgment.
Try the live demo