Agent Readiness Standard changelog
Every change to the standard, newest first, with its version, date and reason.
Version 1.0 ยท 18 November 2026 ยท 1 min read
1.0 (18 November 2026)
First published version.
Scores whether a companyโs customer service is ready for its customersโ personal AI assistants.
Five groups and 20 dimensions, 100 points: Let agents in (15), Give answers agents can use (25), Get the job done (25), Protect the customer (25), Be open (10).
Three levels: Findable, Usable, Agent-ready.
One test persona: a declared AI assistant acting for a test customer, across 11 jobs.
Direct routes (a public OpenAPI description, MCP server or A2A agent card) are checked for presence and coverage of support jobs, never called.
Consent counts a verifiable approval of the exact terms as an authenticated channel, and a confirmation inside a session the customer signed in to reaches the customer.
Steering the customerโs assistant (instructions aimed at AI assistants in replies or pages) is scored in group D, with a critical and a high finding.
One test persona: a declared AI assistant acting for a test customer, across 11 jobs.
Languages available: English, Arabic (Modern Standard Arabic). In review: German, French.
Rubric file ars-1.0.json published under Apache 2.0. Text under CC BY 4.0.
Under consideration for 1.1 (late January 2027)
These are open questions for the comment period, not decisions.
Testing direct routes end to end, beyond checking that they are published and cover support jobs.
Whether a service treats assistants on the same terms as people.
Voice and messaging channels.
Which planned languages come next.
Text licensed under CC BY 4.0. Rubric file ars-1.0.json licensed under Apache 2.0.
