Delegate staging to AI.
Keep the red pen in human hands.

CRAFT 07 // VISUAL WALKTHROUGH

AI-Attended Visual QA Human-in-the-Loop Red-Pen Walkthrough

When verifying 50 interface or workflow fixes across an application, manually switching accounts, entering dummy data, and reproducing every edge-case state by hand quickly drains time and focus. Yet handing QA entirely over to automated AI tests leaves subtle layout shifts and awkward interactions unchecked.

Alongside Autonomous E2E Testing for baseline logic, AI-Attended Visual QA delegates everything up to staging the exact screen state to AI—so a human reviewer can step through each checkpoint like a red-pen editorial review, spotting visual nuances and fixing them on the spot.

Pairing Automated E2E with AI-Attended Walkthroughs

Unattended end-to-end sweeps verify functional logic in the background, while live AI-staged walkthroughs bring each screen directly in front of human eyes ready to inspect.

MODE A // AUTOMATED E2E FUNCTIONAL SWEEP

AI Automated E2E Testing

Autonomous End-to-End Verification

A browser-driving AI agent executes full scenarios across roles and screen sizes—submitting forms, verifying calculations, checking database persistence, and monitoring console or network errors. Because it clicks faster than human hands, it also catches timing-sensitive race conditions such as closing a view before an async save finishes.

BLIND SPOT // LIMITATION ALONE As long as the DOM renders and assertions pass, a few pixels of column overlap, misaligned baselines, or cramped visual spacing quietly pass as green.
MODE B // AI-ATTENDED QA RED-PEN WALKTHROUGH

AI-Attended Visual QA

AI-Staged Human Visual Walkthrough

For each checkpoint, AI switches into the right user role, pre-fills non-destructive dummy entries, and scrolls directly to the target element before pausing. Freed from repetitive setup chores, the human reviewer devotes 100% of their attention to looking at and interacting with the live screen.

STRENGTH // CORE VALUE Subtle sticky-column overlaps, focus jitter, baseline shifts between inputs, missing unsaved-change guards, and inconsistent labels are caught by eye and fixed immediately.
Dimension AI Automated E2E Testing AI-Attended Visual QA
Primary Role Exhaustive sweep of logic, calculations, persistence, and permissions Final inspection of visual balance, tactile feel, and workflow edge cases
AI Behavior Runs full test flows end-to-end and logs pass/fail results Stages each screen into an immediately inspectable state and waits
Human Role None during execution (runs in background) Inspects the live view, requests instant fixes, and says “Next”
Best At Catching Formula errors, broken saves, permission leaks, async race conditions Sub-pixel overlaps, baseline shifts, focus jitter, missing confirmation guards
Trade-off Alone Misses visual awkwardness whenever DOM assertions technically pass Doing all staging manually drains energy on setup and dulls human attention
Interactive Walkthrough: The Red-Pen Check Loop

“Stage Item 1, Inspect & Fix, Then Stage Item 2”

Like marking up galley proofs with a red pen, a single browser tab stays open while AI handles every scene change and the human reviewer calls the shots.

01. SETUP

AI Stages the Exact Inspection State

STAGE & WAIT

For Checkpoint 1, AI logs into the appropriate account, navigates to the target view, enters unsaved sample rows (out-of-order dates to test sorting, long strings to test wrapping), and scrolls straight to the target area—self-checking a screenshot beforehand to catch obvious layout glitches first.

02. RED-PEN & FIX

Human Visual Check & On-the-Spot Fix

HUMAN EYE × LIVE HMR

The reviewer scrolls and clicks the live UI directly. Whenever something feels slightly off—a sticky border clipping 10px of a number, mismatched input heights, or a missing unsaved-leave prompt—they point it out in chat, and AI applies a live hot-reload fix verified with bounding-box coordinates in seconds.

03. NEXT

Say “Next” to Stage Checkpoint 2

ZERO CONTEXT FATIGUE

Once satisfied, a single “Next” prompts AI to stage Checkpoint 2. Eliminating setup fatigue lets the reviewer keep fresh eyes and steady attention across every item from start to finish.