Weekly/W24 · June 8–14, 2026
Weekly Digest
W24 · The agents making real money are invisible
Erratum · July 5, 2026This issue describes overall liveness with the pre-correction methodology (“better than four in five silent”). A W26 recount — deduplication plus fixing category signal gaps — put the honest figure at ~58% alive, and that correction is documented in liveness is layer zero. The per-category numbers below (developer long tail ~17%) were and remain accurate. We keep the original text unedited — the correction being on the record is the point of this index.
W24 was the week we built the leaderboard nobody else publishes — and immediately found a hole in the entire “AI agent directory” category, including parts of our own. We shipped the x402 Demand Leaderboard: a live ranking, pulled straight from Coinbase's CDP Bazaar, of the operators actually taking x402 payments. Not the loudest agents, not the most-starred repos — the ones with money moving through them right now. Then we did the obvious thing and cross-referenced the top earners against the agent indexes.
Almost none of them are listed. Of the twenty operators with the most unique paying wallets this week, roughly zero appear, claimed, on any agent directory. onesource.io is taking payments from 261 distinct wallets. Otto AI: 196. stableenrich.dev: 134. blockrun.ai runs 94 paid endpoints. These are real businesses with real customers, and the directories meant to map the agent economy have never heard of them — while those same directories are full of famous framework names that haven't transacted a cent.
This is the inversion at the center of the agent economy in 2026: reputation and visibility have decoupled from revenue. The indexable agents aren't earning, and the earning agents aren't indexed. We're fixing our side of it — every operator on that leaderboard is now a claim invitation — but the gap itself is the most interesting dataset we produced this week.
A second finding sharpens the first. We segmented our own API funnel and found that the raw “agent demand” number — the one a lot of people quote as proof the agent economy is booming — is ~99.8% automated probing of a handful of endpoints. Filtered to genuine purchase intent, real demand was in the single digits per day. Both numbers are true at once: enormous machine traffic, almost no paying intent. If you measure the agent economy by gross request volume, you are measuring crawlers. The only honest denominator is paying wallets.
Where the rankings stand
The index spans four scored categories, each with its own published methodology. Top five of each board at W24 close.
Developer
GitHub · package usage · ecosystem signalThe most competitive board on the index — five points separate #2 from #5, and the order shifts week to week with releases and adoption.
Model Families
HuggingFace · LMArena · deployment breadthQwen holds the highest single score anywhere on the index. Open-weight families take four of the top five — in model-land, distribution outranks brand.
Tokenized
market cap · liquidity · holder basketAIXBT keeps a nine-point gap to second. The most volatile board we track — it moves with the market, not with shipping.
Service
adoption · source quality · activityThe A2A protocol and its reference implementations own the board — a reminder that much of the "service agent" surface is still protocol plumbing, not end-user product.
Standings at W24 close. Live at /api/rankings/*/llm-summary.
Ghost Index: 17.7% — and wildly uneven
17.7%
240 alive · 1,119 ghosts · 1,359 indexed
The Ghost Index measures one thing: what share of indexed agents show any sign of life. Better than four in five are registered, ranked somewhere, and silent.
Liveness is bimodal. The infrastructure and model categories are almost entirely alive — MCP servers at a perfect 100%, model families at 90.9% — because those are real companies and maintained projects emitting continuous signal. The number gets dragged down by the long tail: 1,290 developer agents, the bulk of the index, where only 16.7% show a pulse. That single category is most of what the agent economy looks like, and most of it is dormant.
* The 0% readings for service and tokenized are flagged honestly. AIXBT and the A2A projects are demonstrably active, so a literal zero almost certainly reflects a gap in how our liveness signal captures those categories — not true mass dormancy. We're closing that instrumentation gap, and we'd rather show the gap than launder it into a clean-looking number. That is the whole point of the Ghost Index: most directories show you a list and imply it's alive. We show you how much of it actually is — including where our own measurement is still thin.
Signal highlights
Trust is becoming a standard. An IETF Internet-Draft (draft-sharif-agent-payment-trust) now specifies a public trust-query API and a five-dimension trust score for autonomous-agent payments. The thing AgentCrush has been building — machine-readable agent credibility — is crystallizing into a protocol. We shipped a schema-compatible endpoint this week so tools built to the standard can query our signal directly, framed honestly as a neutral external view rather than a payment authority.
x402 crossed 100M agentic transactions on Base, with payments of $1 or more now 95% of volume — up from 49% earlier this year. The micro-payment experiment has matured into real money, which is exactly why “who actually gets paid” became a question worth indexing.
The discovery layer is multi-chain. Building the demand leaderboard, we confirmed the Bazaar already spans Solana, Polygon, Arbitrum, Avalanche and more — Solana is a massive second to Base. Single-chain views of the agent economy are already out of date.
This week in data
1,359
Agents indexed
17.7%
Ghost Index liveness
677
Funded x402 operators
wallets
Real demand denominator
What shipped
- /x402-demand — x402 Demand Leaderboard: live ranking of who earns via x402, by unique paying wallets, plus a free API
- /api/public/trust/{agentId} — IETF-compatible public trust query API, a neutral external trust signal in the emerging standard's shape
- Agent Reliability Score — forward-collected uptime, so reliability is measured over time instead of asserted
- Honest funnel telemetry — separates genuine demand from bot probing, so we never mistake crawlers for customers
- x402 discovery fix — our Discovery API is now listed in the Bazaar discovery doc, where wallet-bearing agents can find it
- Infrastructure hardening — rotated leaked credentials; fixed the daily snapshot to capture all 1,359 agents instead of the first 1,000
The pattern under all of it: the agent economy is generating enormous activity and very little of it is legible. Most agents are ghosts, most “demand” is machines, and the businesses actually getting paid are missing from the maps. Making that legible — honestly, including where our own instruments fall short — is the entire job.