Six findings · every one independently reproducible
This brief leads with evidence, not opinion. Every finding below is a public HTTP response you can reproduce from any terminal: a URL, a status code, a content-type, and the exact command. Where we credit Google, that's evidenced too. The EcoWealth positioning at the end rides on the proof; it never runs ahead of it.
Method: passive only. Plain curl GETs of standard public paths (/llms.txt, /ai.txt, /robots.txt, /.well-known/*) plus a soft-404 control path per host to prove the 404s are real, not soft-404 pages. WebSearch / WebFetch for public facts, quoted with source URLs. No authentication, no scanning, no fuzzing, no probing beyond documented well-known paths. Captured 2026-07-14.
| Surface / path | Status | Content-type | Read |
|---|---|---|---|
| www.google.com/llms.txt | 404 | text/html | No LLM index |
| www.google.com/this-is-a-soft404-control-xyz123.txt | 404 | text/html | Control: proves 404s are real |
| www.google.com/.well-known/agent-card.json | 404 | text/html | No A2A agent card |
| www.google.com/.well-known/ai-plugin.json | 404 | text/html | No plugin manifest |
| www.google.com/ai.txt | 404 | text/html | No AI-usage file |
| www.google.com/.well-known/security.txt | 200 | text/plain | Exemplary: credited |
| blog.google/llms.txt · /.well-known/agent-card.json | 404 | text/html | Newsroom: no agent surface |
| sustainability.google/llms.txt · /.well-known/agent-card.json | 404 | text/html | Sustainability: no agent surface |
| gemini.google.com/llms.txt · /.well-known/agent-card.json | 404 | text/html | Gemini host: no agent card |
| developers.google.com/llms.txt · /.well-known/agent-card.json | 404 | text/html | Dev portal: no agent card |
| ai.google/llms.txt | 302 | → /llms.txt/ | SPA catch-all redirect, not a file |
| a2a-protocol.org/llms.txt (Google-created standard) | 200 | text/plain | Real file: the contrast |
Reproduce the whole table: for u in llms.txt ai.txt .well-known/agent-card.json .well-known/security.txt; do curl -s -o /dev/null -w "%{http_code} %{content_type} $u\n" https://www.google.com/$u; done
No jargon. For a company Google's size the value isn't a bill you save; it's whether the agents already reading the web represent Google accurately, and whether the standards and goals Google set can actually be verified. Here's what each fix is worth.
Google created the agent-discovery standard and gave it away, yet an agent that reaches google.com finds no card to read; publishing one makes Google the reference implementation of its own protocol, first where a Gemini user's question actually lands.
Right now an AI describing Google's labor, AI, and sustainability story guesses from prose; a small map lets it answer correctly instead: the same easy, in-character file work Google already does exceptionally for security.
The commitment lives in a report an auditor or procurement counterparty has to take on faith; a per-project, checkable outcome receipt turns an annual attestation into a continuously auditable ledger: the difference between a claim and a proof.
Google ships agents that can act but have nowhere trustworthy to route real-world ecological work; a live endpoint they can discover and pay keyless gives those agents somewhere good to send a person, and returns a receipt.
What we observed
Google launched the Agent2Agent (A2A) protocol in April 2025 and on June 23, 2025 donated it to the Linux Foundation, seeding a vendor-neutral project with AWS, Cisco, Microsoft, Salesforce, SAP and ServiceNow. A2A's own site describes agents that discover each other's capabilities, negotiate interaction modes, and collaborate on tasks.
Yet every Google surface we checked returns 404 for /.well-known/agent-card.json (the A2A discovery file) while a2a-protocol.org/llms.txt returns a real 200 text/plain document.
Proof (reproduce it)
Why it matters to Google
A2A only creates value when agents can find each other. The most credible proof of the standard is for its author to be legible by it, first on the surfaces a Gemini user's question actually lands on. Right now an A2A-speaking agent that reaches google.com gets an HTML 404 where a capability descriptor should be.
llms.txt on any Google surface we surveyedWhat we observed
/llms.txt (the emerging convention that hands an LLM a clean map of a site) returned 404 on google.com, blog.google, sustainability.google, gemini.google.com and developers.google.com. A soft-404 control path on each host also 404s, which proves these are genuine "not found" responses, not a soft-404 page masquerading as content. On ai.google the path 302-redirects to a trailing-slash SPA route (the marketing app's catch-all), i.e. no real file either.
Proof (reproduce it)
Why it matters to Google
Assistants increasingly read llms.txt to decide what a site offers before crawling it. A one-file map on the surfaces that carry Google's labor, AI and sustainability story would let any assistant, Gemini included, answer accurately about Google's commitments instead of guessing from prose.
security.txt is best-in-classWhat we observed
www.google.com/.well-known/security.txt returns a clean, complete, RFC 9116-shaped file: multiple Contact: lines (g.co/vulnz, [email protected]), an Encryption: key, Acknowledgments:, a Policy: (g.co/vrp), a Hiring: line, and Expires: 2030-04-01. Google plainly values structured well-known files and maintains them years out.
Proof (reproduce it)
Why it matters to Google
This is the honest frame: adding an agent-card.json and an llms.txt is the same class of work Google already does exceptionally for security: a short, maintained, machine-readable file. The lift is small and entirely in character.
What we observed
Google's 24/7 carbon-free energy by 2030 goal (Pichai: we are the first major company that's set out to do this
) and the 2026 Environmental Report are published as narrative pages and a linked document. The substance is real and industry-leading: more than 240 agreements to purchase nearly 35 GW of net-new clean energy,
a Gemini prompt footprint down by factors of 33 and 44. But none of it is exposed as a per-hour, per-project machine-verifiable object an agent (or an auditor, or a procurement counterparty) can fetch and confirm.
Proof (reproduce it)
Why it matters to Google
"Carbon-free energy, matched hour by hour" is exactly the kind of claim the market now wants to verify, not just read. A per-project, on-chain outcome receipt turns an annual attestation into a continuously auditable object: the difference between a report and a ledger.
What we observed
Google is shipping agents (Gemini, Agentspace) and the rails they talk over (A2A, and payment conventions). What those agents lack is a funded, proof-verified place to route real ecological work: a destination where "help with climate," "find me local restoration work," or "offset this compute" becomes a settled, receipted action rather than a link.
The destination (all reproducible)
Why it matters to Google
EWP is the funded, proof-verified endpoint an A2A/Gemini agent can discover and pay keyless over x402 (a web-native way for AI agents to pay per use): work-packet $0.10, health-packet $0.05, plus proof-check / tCO2e / method-template / settlement-attest. First fully settled packet on-chain: workId 14, 2026-07-07. It's a reference implementation of Google's own standard, pointed at the outcome Google's sustainability story is chasing.
Three competence ladders, step-one-anyone-today to mastery: every job funded, proof-verified (photo + GPS + method), settled on Base with a tCO2e receipt. A Gemini agent routes the human; EWP settles the work.
Living Building Challenge: regenerative construction tasks, verifiable on site.
Regenerative food & land work, from step one to mastery.
Ecological restoration mapped to tCO2e: the outcome Google's report chases, as a settled object.
Two small, in-character files (an agent card + an llms.txt) make Google's labor and sustainability surfaces answerable by any assistant. One connector routes a Gemini user into funded, proof-verified ecological work that settles on-chain. The destination is already running.
See it running →The ask: proof first, then a small yes
Proof first, ask second. The deliverable is always an on-chain receipt, never a vague "let's talk."
