Capability brief · Independent · Not affiliated with Base

The docs speak fluent agent. The mission page speaks none.

Four findings. The docs surface is credited in full first: it's excellent. Every finding reproducible with curl: exact command, exact status, exact observation. The ask is at the end, and it is small.

Method. Passive recon only: plain HTTP GETs of public pages and standard well-known paths across base.org and docs.base.org, with a nonsense-path control to tell true 404s from soft responses, and body/content-type inspection on every 200. No auth, no fuzzing, no enumeration. Captured 2026-07-14. Mission language cites Base's public 2026 strategy; reported figures labeled as reported.

The bottom line, in plain English

Give base.org the agent surface your docs already nailed → the builders your mission names can actually be found there

An AI that lands on your flagship to answer “who's building on Base?” hits a 404 with nothing to read, so the agent half of “a million builders, human or agent” can't discover the showcase at your front door — the exact audience you're courting quietly bounces.

Put the mission in the machine, not just the copy → agents can act on it, not only read the words

base.org says “builders, agents, onchain” in beautiful human prose, but there's no structured surface an assistant can parse — so a mission written for agents stays something agents can't act on until it exists in a form they read.

You already perfected the pattern one host over → inheriting it is the whole fix

docs.base.org is one of the most agent-legible surfaces online; carrying that same llms.txt-and-agent-card pattern to base.org isn't new invention, it's the small step between promising “human or agent” and being readable by both.

Baseline: credited first, in full

docs.base.org is one of the most agent-legible surfaces on the internet. It ships a real 61 KB llms.txt that names "AI Agents" as a first-class topic, a full llms-full.txt, and a real A2A (a protocol that lets AI agents discover and call each other) agent-card.json (application/json, skill "base", pointing at Base Chain / Base Account / Base MCP). It documents Base MCP ("Give your AI assistant a wallet"), Builder Codes for Agent Developers, and x402 (a web-native way for AI agents to pay per use) payments. This is the pattern everyone else is racing toward, already in production. The findings below are not "Base is behind"; they are "the flagship domain hasn't inherited what your docs team already perfected."
Finding 1 · credit

docs.base.org is fluent in agent: llms.txt, llms-full.txt, and a real A2A card

Wheredocs.base.org/llms.txt · /llms-full.txt · /.well-known/agent-card.json
Statusall 200 · llms.txt 61,120 B, text/plain · agent-card.json application/json, protocolVersion 0.3 · server Vercel
curl -s https://docs.base.org/llms.txt | grep -i 'AI agent\|Base MCP\|x402' | head
> ...Smart Wallet, OnchainKit, MiniKit, Base Chain RPCs, and AI Agents. - [Resources for AI agents](.../get-started/resources-for-ai-agents) - [Base MCP]: Give your AI assistant a wallet. ... pay with x402.

The docs surface tells an agent exactly what Base is, how to get a wallet, and how to pay with x402. Credited without reservation: it's the reference implementation.

Why it matters: it makes the fix trivial. base.org doesn't need to invent anything; it needs to inherit the pattern one host over.

Finding 2 · the headline

base.org, the flagship builder surface, is a true-404 desert for agents

Wherebase.org/llms.txt · /robots.txt · /.well-known/agent-card.json · /ai.txt
Statusall 404 (Next.js __next_error__ shell; a nonsense-path control returns the identical 404: confirmed true 404s, not soft-200s)
# the flagship domain, and the control, return identical real 404s curl -s -o /dev/null -w '%{http_code}\n' https://base.org/llms.txt curl -s -o /dev/null -w '%{http_code}\n' https://base.org/robots.txt curl -s -o /dev/null -w '%{http_code}\n' https://base.org/this-nonsense-control-xyz123.txt
404 (base.org/llms.txt) 404 (base.org/robots.txt: no robots at all) 404 (nonsense control: identical __next_error__ shell)

The domain that carries the mission ("a million builders, human or agent") offers agents nothing machine-readable: no llms.txt, no robots.txt, no A2A card, no ai.txt. An assistant that lands on base.org to answer "who's building on Base and what do they do?" has no structured index to read. The showcase is invisible to the exact audience the mission names.

Why it matters: "human or agent" is a two-sided promise. The human side gets a beautiful rendered page. The agent side gets a 404. Half the intended audience can't parse the front door.

Finding 3

The mission is in the prose, not in the machine

Wherebase.org server-rendered HTML
Observed"agent" ×5, "Builder", "million", "onchain" present in the page text, but no llms.txt / agent card / structured builder index exposes any of it
curl -s https://base.org/ | sed 's/<[^>]*>//g' | grep -io 'agent\|builder\|million\|onchain' | sort | uniq -c
5 agent 1 Builder 1 million 3 onchain

base.org says the words (builders, agents, a million, onchain) in human-readable copy. But saying "agent" in prose and being readable by an agent are different things. There is no llms.txt to index the message and no agent card to make the showcase's builders callable or discoverable. The narrative exists; its machine-readable twin doesn't.

Why it matters: agents don't read marketing prose the way humans do; they read structured surfaces. A mission addressed to agents that only exists in prose is a mission agents can't act on.

Finding 4 · the worked example

Here's exactly the "human or agent" builder your mission wants, already live on Base

EcoWealth runs a complete real-world-work settlement loop on Base mainnet: a place + problem becomes a bounded work packet → funded → claimed (by a human or an agent) → proof (photo + GPS + method signature) → approved → settled on-chain → retirement receipt where ecological. It is the full physical-economy flywheel, on your chain.

On-chainEWP contract 0x76c17C…A14B: source-verified on BaseScan; first settled packet workId 14
Board7,624 postings / ~$671k posted value (vealth.net/labor/stats)
Agent-payablestandard x402 at vealth.net/.well-known/x402 (22 paths, USDC on Base), the same x402 your docs teach

Why it matters to Base: this is the builder your mission is written for, and today base.org can't surface it to an agent. Extend the docs.base.org agent pattern to base.org and builders like this one become discoverable to the million agents you're upgrading the chain for.

The ask

Proof first, then one small step tied to your own million-builders mission. The deliverable is an on-chain receipt on Base you verify yourself.

Showcase one live builder. Index EcoWealth's real-world-work loop as a reference "human or agent" builder on base.org, already on Base mainnet, no changes to Base systems.
Add the surface it needs. Extend the docs.base.org pattern to base.org: a builder-discovery llms.txt + agent card. Concept draft in the Agent Kit.
Fund one packet. Back one proof-verified ecological packet ~$10 → the on-chain receipt, a worked example of the physical-economy loop your mission names.
← Back to the cover
Provisioned by EcoWealth · Digital Provisioning Protocol · provenance‑sealed