Answers are grounded in this paper and cite their sources.
Software agents are starting to search, compare and buy on behalf of people and companies, but they can only consider suppliers they are able to find and read. This working paper measures whether 200 business-to-business (B2B) companies in six Spanish-speaking Latin American countries (Argentina, Mexico, Colombia, Chile, Peru, Uruguay) publish the public web infrastructure that agents use for discovery, and compares them with the ten AI vendors measured in the first edition of the benchmark. It introduces PARS v2, a pre-registered 100-point score in which 90 points are computed automatically from raw HTTP observations, including conformance of Agent Cards to A2A 1.0, signature verification and endpoint liveness. Of 177 companies that a standard HTTP client could read, none published an Agent Card at the standard path; the median score was 6 out of 90. The AI vendor panel scored the same median. Readiness was concentrated in technology and fintech companies. A further 11 companies blocked automated clients and 9 served TLS certificates that standard clients reject. This record contains the paper, the sample, raw HTTP observations for the pilot and the two collection waves (1 October 2026), PARS v2 scores, the coding of dimension H with evidence, the comparison with Cloudflare's Agent Readiness checker, and all code (crawler, scoring and pre-registered analysis). Data: CC BY 4.0. Code: MIT. Pre-registration: https://doi.org/10.5281/zenodo.23088674
No takes yet. Share an insight, caveat, or question.
Gabriela de Marco (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: