Agent Readiness Across 1,157 Vibe-Coded Project Pages
90.9% of measured VCL project pages received Cloudflare Agent Readiness levels L0 or L1. The full distribution and the limits of this provider-defined benchmark.
1,052 of 1,157 measured public project pages—90.9%—received L0 or L1 on Cloudflare’s Agent Readiness scale in VCL’s September report. L1, labelled “Basic Web Presence” by the provider, was the most common result.
That is evidence of low recorded levels under this particular benchmark. It is not evidence that 90.9% of products are unusable by every AI agent.
The six recorded levels
VCL submitted project URLs to Is It Agent Ready and retained the provider’s returned level. Here are the provider-defined labels used in the edition:
| Recorded provider level | Project pages | Share of 1,157 measured pages |
|---|---|---|
| L0: Not Ready | 412 | 35.6% |
| L1: Basic Web Presence | 640 | 55.3% |
| L2: Bot-Aware | 87 | 7.5% |
| L3: Agent-Readable | 7 | 0.6% |
| L4: Agent-Integrated | 4 | 0.3% |
| L5: Agent-Native | 7 | 0.6% |
The denominator is 1,157 measured URLs, not the full 1,254-project cohort. The 97 projects without a qualifying measurement are excluded rather than assigned L0. The headline combines the recorded L0 and L1 counts: 412 plus 640.
A scanner level is not an agent task-completion rate
Agent Readiness is not a universal standard. These levels cannot be directly compared with another provider’s percentage score simply because both use “agent-ready” language.
The report does not measure whether an agent successfully bought a product, completed onboarding or used an authenticated feature. A low level is also not proof that search engines cannot index the page. Those would be different tests.
The target matters too. A project submitted with an app-store or marketplace URL contributes a measurement of its listing page, not of the underlying app. Each result is the latest successful recorded measurement before the cutoff; the scans need not have happened on the same day.
Within those boundaries, the distribution is clear: most measured pages were concentrated in the provider’s first two levels. A stronger claim about practical agent access would require task-based testing on the applications themselves.
Data and source. State of Vibe Coding 2026, Dataset v7: Agent Readiness, cutoff September 11, 2026 at 09:40:56 UTC. See the edition’s methodology for provider and scan-target details.
The full State of Vibe Coding report compares this distribution with other, separately defined public-web measurements.