The Agent Usability Directory

What it is

A directory at /usability-directory. Every company on the page is a real company, scored against the Agent Usability Report rubric. You can sort by highest score, lowest score, name, or most recently scored. You can search by company name or domain. You can press ⌘K anywhere on the site to jump straight to a specific company’s scorecard — or to kick off a fresh scan if it isn’t in the directory yet.

It is not a “top startups” list or a curated awards shortlist. There is no editorial judgement. The score is the score.

Why it exists

The Agent Usability Report answers one question well: how agent-ready is this single URL? The directory answers the question a step up: who in the market is actually agent-ready?

Operators want to know who their peers are. Investors want a quick filter for “agent-native” vs “agent-curious” vs “agent-blind.” Builders want to know which of the companies they integrate with will give an agent a usable interface — and which will quietly break inside their workflow. A leaderboard against a stable rubric makes that legible. One number per company, comparable across the entire field.

It also keeps us honest. Anyone can game a single demo URL. Scoring 900+ companies on the same checks, with deterministic logic, means the rubric earns its weight in public.

How it works

Every company on the page has been through the same pipeline:

1. We fetch the homepage with a labelled agent user-agent and a real browser.
2. We probe roughly 50 conventional paths — /llms.txt, /robots.txt, /sitemap.xml, /.well-known/*, /api, /docs, response headers, HTML semantics.
3. The pipeline aggregates those signals into the same five categories scored in the report: discovery, APIs & agent endpoints, content & semantics, commerce & reliability, security & trust.
4. The score is locked. The verdict and explanation are written after. Two scans of the same site produce the same number.

No LLM is involved in scoring. Cost per scan is effectively zero, so we can run the long tail without thinking about budget.

Re-runs land in the directory automatically. If you scan a site through the public form, it shows up here once the scan completes. The cache is annual — re-submitting an already-scanned domain returns the existing report rather than burning another scan unless you explicitly want a fresh one.

How to read the directory page

At the top: four numbers — websites ranked, average score, AI ready (score ≥ 70), needs work (score < 70). Those four are the most honest summary of the state of the agent-readable web today, refreshed every time a scan completes.

Underneath: the top 5 by score. Logo, name, what they do, and the five category bars from the rubric. This is the leaderboard view — the companies most operators should look at if they want a benchmark for what “good” looks like.

Then the full list, paginated 50 at a time with infinite scroll. Each row is a link to that company’s full scorecard at /usability-report/{domain} — same rendering as the public scan page, with the full check-by-check evidence trail.

Sorting:

• Highest score — the leaderboard view. Who’s doing this best.
• Lowest score — who’s currently agent-invisible. Useful if you’re a vendor pitching agent-native UX work.
• Name — alphabetical. Useful for ⌘F.
• Most recently scored — what just landed. Useful if you’re watching the directory grow.

⌘K — the fast path

From anywhere on the site, press ⌘K (or Ctrl-K). The palette searches across every company in the directory and surfaces matches as you type — by name, by domain, or by description. Enter opens the scorecard.

If you type a domain that isn’t in the directory yet, the top result becomes “Run a new scan on <domain>”. Pressing enter kicks the scan off. When it finishes, the company shows up here automatically.

This is the primary way to use the directory once you know what you’re looking for. Skip the table; type the name; hit enter.

What it doesn’t tell you

• Whether the product is good for humans. A site can score 0 here and be the cleanest GUI on the internet. This rubric measures agent-readability, full stop.
• Whether the company is well-run, well-funded, or well-loved. It measures the web surface, not the business behind it.
• Whether your agent will succeed. The rubric is about table stakes — can a generic agent discover, call, credential, and operate? A score of 90 makes that possible. It doesn’t guarantee that your specific workflow will work end-to-end. For that, run the scan and read the evidence trail.

Get a company on the directory

If a domain isn’t here yet, scan it. Either submit it on the Agent Usability Report page or open ⌘K and type the domain. The directory updates the next time the page is built, and the per-company scorecard is available immediately after the scan completes.