What you get

Every Agent Design Check engagement is built on the same foundation: QA you can verify. Here's exactly what a working relationship with us looks like.

Recorded QA sessions

Each QA pass is a session run against your app — staging, a deploy preview, or production — and captured end-to-end:

• Screen recording of the full pass, reviewable at any time.
• Screenshots linked to checklist items, so "tested" means demonstrated.
• Notes pinned to the exact position and video moment they describe — no ambiguous "on the settings page somewhere."
• Multi-viewport coverage: desktop, tablet (1024), mobile (390), and TV (1920) breakpoints.

Checklist-driven coverage

Coverage is defined by checklists we build together from your critical flows — sign-up, checkout, dashboards, permissions, whatever your business depends on. Checklists are living documents: they grow with your product, and every session reports completion against them. Coverage gaps are visible, never silent.

Findings, delivered where you work

Nothing lives in a portal you'll forget to check:

• Slack / Microsoft Teams — session summaries and high-severity alerts.
• GitHub / Linear / Jira / Notion — tracked issues with reproduction steps, severity, viewport, and the supporting screenshot or video clip attached.

Frontier-model integrations

If your team builds with Claude, GPT, or Gemini-based coding agents, QA findings plug straight into that workflow. Every finding ships in a structured, machine-readable format written for agents — element, state, viewport, expected vs. actual, and an evidence link — so your agent picks up the bug, sees the proof, and opens the fix without a human translating in between.

The findings format

Every finding we file carries the same fields, whether it lands in Jira or in front of a coding agent. When your team runs agents, we hand over the same finding as JSON:

{
  "id": "ADC-0412",
  "severity": "high",            // high | med | low
  "area": "checkout",
  "title": "Promo code field accepts input but Apply does nothing",
  "viewport": "mobile",          // desktop | tablet | mobile | tv
  "viewport_width": 390,
  "element": "button[data-testid=apply-promo]",
  "state": "logged in, cart with 2 items, code SAVE10 entered",
  "steps": ["Open /cart", "Enter SAVE10", "Tap Apply"],
  "expected": "Discount line appears; total updates",
  "actual": "No network request; total unchanged; no error shown",
  "evidence": {
    "screenshot": "https://…/s/2f9c/0412.png",
    "video": "https://…/s/2f9c#t=00:04:31",
    "checklist_item": "Checkout › Promo codes"
  },
  "status": "verified"           // draft | verified | filed — a human verifies before filing
}

Element, state, viewport, expected vs. actual, and an evidence link — enough for a human to reproduce it in a minute and for an agent to open the fix without asking. The full JSON Schema is published at qa.agentdesigncheck.com/api/v1/findings/schema.

AI-assisted review, human-verified

Our own review is AI-amplified too: session artifacts are additionally analyzed with frontier models to surface design-level issues — contrast and accessibility problems, weak visual hierarchy, inconsistent spacing, confusing flows — and every AI-surfaced finding is verified by a human before it reaches your tracker.

Cadence and turnaround

• Standing coverage — per release, nightly, or weekly, on the cadence we agree.
• Ad-hoc passes — a release candidate or a risky branch, typically returned within one business day.

Getting started

Contact us with what you're shipping. We'll scope coverage together on one short call, agree on cadence and pricing, and your first recorded session usually runs within the week. Wondering how this differs from a QA-as-a-service platform or a managed automation vendor? We wrote it down.