A respectful, opportunity-framed read of Meta's public surface: what an AI agent (including a Llama-based one) sees today, and where a funded, proof-verified, on-chain-settled work layer plugs into Community, Llama, and the water-positive commitment. Every finding is backed by a response Meta's own servers returned to a normal GET.
X-Frame-Options: DENY, X-Content-Type-Options: nosniff, and HSTS with preload. Email is locked down: _dmarc.meta.com and _dmarc.facebook.com both enforce p=reject. And the product substance is real: Llama 4 is the most widely deployed open-weight model ecosystem in the world, and the water-positive-2030 program returned 1.6B gallons in 2024. This brief credits all of that first, then shows where the machine-legibility hasn't caught up.
No jargon. Meta already ships the strongest security and the most-deployed open models we've seen; every fix below just lets the machines catch up to that lead. Here is what each one is worth.
The redirect is already wired; it just lands on a 400 error page, so the company that shipped the world's most-deployed open models has no working AI front door: the lowest-effort, highest-symbolism fix on this list.
Facebook Groups organize cleanups and mutual aid at planetary scale, but the record is just a photo in a feed; a funded, proof-checked, settled receipt turns that same energy into something sponsors will actually fund.
Meta AI runs on Llama across apps used by 3B+ people, but those agents have nowhere to fund and prove physical work; a keyless endpoint gives them a destination to act on, no custody or chain for Meta to build.
The 1.6B gallons returned and the net-zero-2030 target live only as prose and PDF; structured facts get quoted accurately instead of paraphrased from scraped copy.
The intent is already wired: llama.com/llms.txt issues a 301 to a canonical AI-docs path. But that destination returns a 400 error page, not a markdown file. So the company that gave the world its most-deployed open models has no agent-readable llms.txt that actually resolves, anywhere.
This is the lowest-effort, highest-symbolism fix we've found: the redirect already points somewhere, the endpoint just needs to serve a real file instead of a 400. For the standard-bearer of open AI, an agent-readable front door is table stakes.
Facebook Groups is where cleanups, mutual aid, tree plantings, and disaster response are organized at planetary scale. But the record of the work is a post, a photo in a feed. There is no funded, claimable, proof-checked, settled object behind "we did it." (This is a structural observation about the product surface, not a defect claim.)
EWP is exactly the missing layer: place + problem → bounded work packet → funded → claimed → proof (photo+GPS+signature) → settled on-chain. A Group's "community action" becomes a verifiable outcome its members and sponsors can cite, the same energy Groups already generate, now with a receipt. It turns organizing into settled work without Meta building custody or a chain.
Meta AI runs on Llama across WhatsApp, Instagram, Facebook, and Messenger (apps used by 3B+ people) and open-weight Llama agents proliferate across the ecosystem. What an autonomous agent lacks is somewhere to fund and settle real-world work keylessly.
EWP packets are keyless-payable over standard x402 (a web-native way for AI agents to pay per use) by any agent, including a Llama-based one, no facilitator, no custody by Meta. Meta ships the brains to the world; EWP is a destination those brains can act on and prove. It makes Llama agents useful in the physical world, with the receipt on a public chain.
Meta commits to net-zero across its value chain in 2030 and to being water-positive in 2030 (restore 200% of consumption in high-water-stress watersheds; 40+ projects, 1.6B gallons returned in 2024, 23.8M tCO2e reduced). The reporting lives on a real site, but it's prose/PDF, with no agent-readable structured discovery.
Water-positive restoration (200% in high-stress watersheds) is precisely the kind of place-based ecological outcome EWP settles with photo+GPS+method receipts. A settled gallon or ton is worth more than a reported one, and it's the outcome layer under the number nobody in the peer set has.
A ~6-line security.txt pointing at the existing whitehat program closes a small discoverability gap for automated researchers. Offered only because "ruthless" means honest: the rest of the security baseline is among the best we've seen.
Phase 0, today, no Meta changes: an Agent Kit over Meta's public data on EcoWealth infra + one Llama-based agent making a live x402 call, and one water-restoration packet funded on Base in a named watershed (the built demo). Phase 1, pilot: serve a real llms.txt at the endpoint the redirect already targets; expose a priced agent-card.json Llama agents can read; light up a proof-of-outcome lane under one water/climate line item. Phase 2, at volume: Community/Groups action and a sustainability budget fund packets; each becomes a settled, citable outcome.
| Claim | Status / value | Re-verify |
|---|---|---|
| llama.com/llms.txt redirects to a 400 | 301 → developer.meta.com/ai/llms.txt/ = 400, 379 KB HTML | curl -sL -o /dev/null -w '%{http_code} %{url_effective}\n' https://www.llama.com/llms.txt |
| No llms.txt resolves on any Meta surface | meta.com/about/ai.meta/facebook = 400; sustainability.atmeta = 404 (controls match) | for u in www.meta.com about.meta.com ai.meta.com www.facebook.com sustainability.atmeta.com; do curl -s -o /dev/null -w "%{http_code} $u\n" https://$u/llms.txt; done |
| Strong security headers (credited) | CSP nonce + XFO DENY + nosniff + HSTS preload | curl -sI https://www.meta.com/ | grep -iE 'content-security|x-frame|strict-transport' |
| DMARC enforced p=reject (credited) | p=reject on meta.com & facebook.com | dig +short TXT _dmarc.meta.com ; dig +short TXT _dmarc.facebook.com |
| Sustainability site real but no llms.txt | robots 200 + sitemap; llms.txt 404 | curl -s https://sustainability.atmeta.com/robots.txt ; curl -s -o /dev/null -w '%{http_code}\n' https://sustainability.atmeta.com/llms.txt |
| No standard security.txt | /.well-known/security.txt = 400 | curl -s -o /dev/null -w '%{http_code}\n' https://www.facebook.com/.well-known/security.txt |
| Llama 4 = most-deployed open-weight ecosystem | public record, Apr 2026 | WebSearch "Llama 4 most widely deployed open-weight model 2026 Meta AI" |
| Water-positive 2030 (200% high-stress; 1.6B gal 2024) | sustainability.atmeta.com/water | WebSearch "Meta water positive 2030 200% high-stress 1.6 billion gallons 2024" |
| EWP live on Base mainnet | contract 0x76c17C…A14B, workId 14 | BaseScan the contract; curl -s https://vealth.net/.well-known/x402 |
Not "let's talk." Pick one small step; the deliverable is an on-chain receipt, not a meeting. Both are pocket change against ~$160B+ revenue and tie straight to what you already ship and promise:
$0.10: point one Llama-based agent at our live work-packet endpoint (one keyless x402 call) and watch a real bounded packet come back. You ship the open brains; give one a place to act. Reproducible now: curl https://vealth.net/.well-known/x402.
$50: fund one proof-verified water-restoration packet in a data-center watershed you already report on. It returns a settled photo+GPS+method receipt on Base: one on-chain verified outcome under the water-positive-2030 number, in one afternoon.
Then fifteen minutes to look at the result in Meta's palette, no system access, nothing probed. Everything in this brief was read from your own public responses.
