ora // blog
How ready is the web for agents?
What we're learning from scanning thousands of products for how well AI agents can find, understand, and use them.
Deep Scan v2: we ran the agents, then scored what they did
The old scan sent agents to role-play our checklist on every run. It was our opinion, and it was slow. v2 flips it: we run real agents across thousands of intents and sites in our lab, learn where they succeed and stall, and reverse-engineer the score from that. The live scan now finishes in 20 to 30 seconds, and only signals we have verified agents rely on count.
What agents actually reach
Real agents lean hard on the human web - homepages and docs, in the same order a person would - and barely reach the files we built for them. Where those files do get reached, they convert better than anything. The bottleneck is discovery, not merit.
Auth is the last mile of agent autonomy
Every layer of agent readiness improves each quarter. Auth doesn't. The sign-up funnel was built for humans with browsers, mailboxes, and patience, and an agent has none of those. We walk the full surface an agent negotiates today - API keys, OAuth and PKCE, token exchange and multi-hop delegation, client registration, Web Bot Auth, ID-JAG, and the x402 payment bypass - where each one stops, and what Deep Scan v1.2 now grades across the Access layer. Plus a perf rewrite that takes a full scan from ~120s to ~15s.
Deep Scan v1.1: every layer just got deeper
v1 ran real agents at every layer. v1.1 makes those agents work harder. Discovery is now a true AEO/GEO benchmark across answer engines. MCP gets graded on Anthropic's own best-practice guidelines, not just whether the endpoint responds. Some scores will drop. That's the point.
The state of agent readiness
Agents are the new customers. We scored thousands of products on whether agents can actually work on their sites. 99% aren't ready. Here's what the 1% are doing differently.
Introducing Deep Scan: a benchmark, not a checklist
Static scanners can tell you whether your site serves the right files. They can't tell you whether an agent picks you. Deep Scan spawns real agents at every layer of the AgentReady standard, from open-web discovery to multi-turn task completion, and grades on what they actually did.
Introducing AgentReady: the first open standard for agent readiness
The agentic web has a dozen protocols and no shared way to name what a product should implement. AgentReady is the first open standard for agent readiness: vendor-neutral, MIT-licensed, and deliberately separate from how anyone scores against it.