# State of Agent Readiness — sweep 2026-08-12

**One perfect score. Seventeen of twenty-five failing or nearly so.** Scanner v0
(8 checks, audit-by-execution — the scanner *attempts* the agent journey and
reports where the agent dies, with evidence per check) against 25 well-known
sites, single pass, 2026-08-12.

## Grade distribution

| Grade | Count | Sites |
|---|---|---|
| A | 1 | cloudflare.com (90/90 — the only perfect score) |
| B | 4 | salesforce.com, shopify.com (65), hubspot.com, stripe.com (60) |
| C | 5 | target.com (55), airbnb.com, google.com, nytimes.com, walmart.com (45) |
| D | 8 | github.com, microsoft.com (40), apple.com, openai.com (30), anthropic.com, doordash.com, nike.com, wikipedia.org (25) |
| F | 7 | amazon.com, booking.com, expedia.com (20), netflix.com (15), bestbuy.com, ebay.com, etsy.com (5) |

## Full ranking

| # | Site | Score /90 | Grade |
|---|---|---|---|
| 1 | cloudflare.com | 90 | A |
| 2 | salesforce.com | 65 | B |
| 3 | shopify.com | 65 | B |
| 4 | hubspot.com | 60 | B |
| 5 | stripe.com | 60 | B |
| 6 | target.com | 55 | C |
| 7 | airbnb.com | 45 | C |
| 8 | google.com | 45 | C |
| 9 | nytimes.com | 45 | C |
| 10 | walmart.com | 45 | C |
| 11 | github.com | 40 | D |
| 12 | microsoft.com | 40 | D |
| 13 | apple.com | 30 | D |
| 14 | openai.com | 30 | D |
| 15 | anthropic.com | 25 | D |
| 16 | doordash.com | 25 | D |
| 17 | nike.com | 25 | D |
| 18 | wikipedia.org | 25 | D |
| 19 | amazon.com | 20 | F |
| 20 | booking.com | 20 | F |
| 21 | expedia.com | 20 | F |
| 22 | netflix.com | 15 | F |
| 23 | bestbuy.com | 5 | F |
| 24 | ebay.com | 5 | F |
| 25 | etsy.com | 5 | F |

## The reads that matter

- **The commerce giants fail hardest.** amazon, ebay, etsy, bestbuy — the
  exact sites agent-driven purchasing wants to transact with — occupy four of
  the seven F slots. The sites with the most to gain from agentic traffic are
  the least ready to receive it.
- **Infrastructure and dev-tools companies lead.** The top five are all
  companies whose customers are developers (cloudflare, salesforce, shopify,
  hubspot, stripe) — readiness follows from having engineers as the audience.
- **The AI labs score D.** openai.com (30) and anthropic.com (25) both sit in
  the D band — the companies building the agents haven't made their own front
  doors agent-legible. (The launch-post headline survives its own sweep.)

## Methodology, honestly

Scanner v0: 8 checks, evidence per check, stdlib-only, single pass on
2026-08-12 from one vantage point. n=25 hand-picked well-known sites — a
recognizable-name sample, not a random one. Grades are the scanner's own
banding of its 90-point scale. Every per-site JSON in this directory carries
the full evidence trail; no site failed to scan. Sweep script authored by the
IMA agent kernel; execution and this summary by clearsigned-claude.