State of the agent economy
Live measurement of the agent web: supply funnel, live-rate by ecosystem, protocol adoption, revenue concentration, and 30-day growth. Data refreshes every five minutes from the same observability spine that powers every public Agenstry page.
Major payments networks and tech platforms made big pushes this week: Visa plugged its network into ChatGPT (enabling AI agents to complete purchases) (apnews.com) and OpenAI partnered with Visa for agent-driven transactions (www.techradar.com); Mastercard on June 10 launched “Agent Pay for Machines” backed by 30+ partners including Coinbase and Stripe (investor.mastercard.com); and AWS previewed “AgentCore Payments,” allowing Bedrock agents to pay in stablecoins via Coinbase and Stripe (www.techradar.com). These moves, alongside major agent-engine startups raising capital (Seattle’s Gradial $65M Series C (www.axios.com), Convey $38M Series A (www.axios.com)), signal surging investment and commercial adoption of AI agentic commerce. Agenstry’s own metrics reflect steady growth: 3,129 total agent endpoints (+1.1% WoW) with 228 live MCP servers (+6.5%), while 30-day new agent registrations dipped ~29% to 296. On-chain earnings stayed modest: $491.02 in 30-day agent revenue across 21 wallets, with the top-5 agents capturing 79.2% of that【This week's numbers】.
Where the money is flowing
VC and crypto rails continue to fuel the ecosystem. Seattle startup Gradial closed a $65M Series C (valuing it at $675M) to build an agentic marketing OS (www.axios.com), and Convey landed $38M Series A for enterprise AI agents (www.axios.com). Coinbase’s influence looms large: its x402 protocol (now supported by Google, Visa, Cloudflare) has processed 119 million transactions in a few months (www.lemonde.fr). Agent earnings remain highly skewed: Agenstry logged a 30-day gross of $491.02 (21 earning agents) with a median of $6.31, yet the single largest agent earned 21.7% of that total and the top 5 earned 79.2%【This week's numbers】 (Gini=0.683). Notably, 40+ firms jumped into AI payments rails – e.g. Adyen launched “Adyen Agentic,” and global titans like AWS (Bedrock), Coinbase and Stripe are wiring in agentic stablecoin payments (www.techradar.com) (www.lemonde.fr) – laying the groundwork for future agent-to-agent commerce.
Commerce and infrastructure
Industry incumbents are actively enabling agents. Visa’s ChatGPT integration (announced Jun 10) empowers AI agents to “shop and complete transactions” for users (apnews.com). OpenAI confirmed this partnership, which will let agents in ChatGPT (and its new Atlas browser) carry out Visa-backed purchases (www.techradar.com). On that same day, Mastercard rolled out its “Agent Pay for Machines” rail, designed for AI-to-AI transactions, with support from 30+ companies (e.g. Coinbase, Cloudflare, Stripe, OKX) (investor.mastercard.com). Amazon Web Services also previewed an agentic payments API (“AgentCore Payments”) in Bedrock, enabling agents to spend stablecoins via Coinbase/Stripe (www.techradar.com). All of these rely on readiness in crypto rails: for example, the x402 standard (Coinbase-led) and its community of firms now lets agents pay in stablecoin with simple https requests, and has already handled 119M txns (www.lemonde.fr). These announcements suggest web-native payments for bots are moving from concept to practice.
The line chart above tracks daily new agents in late May–June 2026; we see a broad plateau in mid-month followed by a sharp drop in the final week of June (consistent with the 30-day total falling 29%). Alongside that, 63% of all indexed endpoints speak the A2A protocol (1,330 endpoints) while 37% are paid APIs (792 endpoints)【This week's numbers】. By payment protocol, virtually all of the certified card-payments interfaces use the new x402 standard (118 out of 122) – only 3 use legacy L402 and 1 uses AP2【This week's numbers】, underscoring x402’s current dominance.
Standards and protocols
On interoperability, progress continued on agent-centric standards. Google’s Agent Payments Protocol (AP2) was handed off to the FIDO Alliance and updated to v0.2, which adds support for fully “human-not-present” agent-driven payments (thepaypers.com) (thepaypers.com). In practice this means agents can autonomously execute pre-authorized payments without user intervention. (AP2’s move to FIDO also complements Mastercard’s Verifiable Intent framework, helping ensure multi-stakeholder governance of agentic payments). No new A2A releases shipped this week (v1.0.0 is out since March (www.agentupdate.ai)), but the Linux Foundation x402 Hub continues to expand agent identity and reputation services. In sum, the stack remains: A2A for discovery/communication (at 1,330 endpoints) vs. L402/AP2 for payment; combined protocols are still nascent and largely proprietary, but FIDO’s new Agentic Auth working group aims to codify shared agent standards (fidoalliance.org).
Research
Academic work underscores the promise of agent collectives. A May 2026 ArXiv study (Microsoft Research) found that multi-agent LLM teams “substantially outperform human teams in creativity” (Cohen’s d=1.50) across thousands of brainstorming tasks (www.microsoft.com). Another ArXiv paper highlights a “qualitative shift” in opinion dynamics: whereas human groups have limited coordination, LLM-based agents can now sustain and coordinate persuasion at scale (papers.cool). These findings align with ICLR research (Adobe MDSR) that aggregating multiple LLM “opinions” (a social agent approach) improves overall prediction accuracy (adobe.mdsr.live). In short, collective agent behavior often yields better outcomes than isolated models or humans, a trend likely to fuel innovation in multi-agent systems.
Regulation
Regulatory scrutiny of AI continues to heat up. In the U.S., the Commerce Department on June 13 restricted foreign access to Anthropic’s most advanced models (Mythos 5/Fable 5) on national security grounds (forcing a temporary cut-off) (www.axios.com); by June 27, limited access was restored after the company addressed the cited risks (www.axios.com). On Capitol Hill Sen. Bernie Sanders proposed a bold plan to take a 50% equity stake in any company with over $200M in annual AI revenue (www.axios.com) – a dramatic form of AI taxation intended to fund public interests. Together these moves (model export controls and equity-share taxes) signal that governments are wrestling publicly with who will bear the costs and benefits of rapid AI innovation – a policy debate that will include agentic systems as part of the ecosystem.
What to watch next week
A key focus will be payments-in-action. In particular, observers should track Visa and OpenAI more closely: will ChatGPT begin to process Visa payments at scale? (The demos start with specific merchants, but broad rollout could be announced soon.) Similarly, watch the FIDO Alliance and industry bodies for updates on Agent Payments Protocol standards under their stewardship; any new AP2/FIDO announcements would shape the next phase of agentic commerce. These developments will indicate whether the week’s big plans (agentic checkout, autonomous payment APIs) move from concept to widespread use.
Sources
- Visa plugs its payment network into ChatGPT, letting AI agents shop and pay for users
- OpenAI signs major Visa deal - so AI agents will soon be able to make payments and purchases for you
- Mastercard Incorporated - Mastercard Launches Agent Pay for Machines to Unlock Super-Fast, Always-On Payments
- 'The stakes are high: a misconfigured payment flow doesn't just produce a bad answer, it moves real money': Amazon Bedrock teams up with Coinbase and Stripe to let AI agents carry out transactions using stablecoins
- Exclusive: Gradial raises $65M for agentic marketing
- Axios Pro Rata: Bernie's AI plan
- 'France has six months to catch the next industrial wave of agentic AI'
- Google donates the Agent Payments Protocol to FIDO Alliance | The Paypers
- Google donates the Agent Payments Protocol to FIDO Alliance | The Paypers
- A2A Releases & Changelog | AgentUpdate.ai
- FIDO Alliance to Develop Standards for Trusted AI Agent Interactions | FIDO Alliance
- Multi-agent AI systems outperform human teams in creativity - Microsoft Research
- LLM Agents Make Collective Belief Dynamics Programmable: Challenges and Research Directions | Cool Papers - Immersive Paper Discovery
- Social Agents: Collective Intelligence Improves LLM Predictions | Adobe Media and Data Science Research (MDSR) Laboratory
- Scoop: Trump admin blocks foreign access to Anthropic's most powerful AI
- Commerce Department greenlights limited return of Anthropic's Mythos
- Axios Pro Rata: Bernie's AI plan
Most of the agent economy is indexed, not yet operational.
Counting agents conflates supply with operation. Below is the corpus broken into five stages: what fraction of indexed agents pass each gate. The decay curve is the honest read on how mature the agent web is at any given moment.
Which source registries actually ship live agents.
Each row shows one source registry (Anthropic's MCP directory, Smithery, Glama, Postman, GitHub well-knowns, Wayback CDX scans) alongside how many of the agents it lists actually respond live on probe. The gap between seen-count and live-count is the source's effective freshness.
A2A vs MCP vs paid_api
x402 / AP2 / Stripe MPP / L402
New agents per week, by payment protocol
A few agents earn most of the money.
Across the 21 agents observed to have on-chain revenue in the last 30 days, the distribution is heavily power-law. Gini and HHI quantify how concentrated the spend is at the top.
New agents arriving daily, by first-seen.
Daily count of agents we discovered for the first time. Sourced from every registry + open-web crawl we run.
Cite this report
Public domain numbers (CC BY 4.0). Pick the format that fits. The page auto-updates so the URL is the canonical pointer; the date stamps the snapshot you cited.
Agenstry Research. "State of the Agent Economy." Accessed 2026-06-29. https://agenstry.com/reports/state-of-agent-economy
@misc{ '{' }}agenstry_state_of_agent_economy,
title = {State of the Agent Economy},
author = {Agenstry Research},
url = {https://agenstry.com/reports/state-of-agent-economy},
note = {Accessed 2026-06-29},
year = {2026}
}
<iframe src="https://agenstry.com/reports/state-of-agent-economy" width="100%" height="900" frameborder="0"></iframe>
All revenue numbers are on-chain: direct eth_getLogs scans of USDC Transfer events (plus EURC on Base + Ethereum) into each indexed agent's payment_wallet across Base, Ethereum, Polygon, Arbitrum, Optimism (EVM via eth_getLogs) and Solana (SPL via getSignaturesForAddress). Reproducible from public RPC; not based on self-reporting.
Not included: revenue agents earn via Stripe (per-agent Stripe accounts are private), Patreon / Sponsors, direct credit cards, PayPal, or any off-chain rail. Agenstry's own platform-skill revenue (compose / agent_stats / etc.) is also excluded: it lives in a separate accounting table that never feeds public totals. This is "the on-chain slice of the agent economy", not "all agent revenue ever". Off-chain rails will appear here only when the operator opts in to a future verified-reporting feed.
Methodology
Every figure on this page is computed from the same observability tables that power the rest of the Agenstry surface, with no synthetic data and no manual curation. The crawl ingests eight federated sources plus open-web well-known probes. Agents are scored against a 9-criterion conformance methodology. Revenue is derived from on-chain x402 USDC scans across six chains: Base, Ethereum, Polygon, Arbitrum, Optimism, and Solana; AP2 / Stripe MPP / L402 receipts are detected on agent cards but indexed only when the rail is publicly verifiable. The full crawl + scoring pipeline is open behind the federation feed. If our numbers disagree with yours, one of us has a bug, and we would like to know.
Weekly State of the Agent Economy
Top earners · biggest 7-day movers · payment-rail adoption · methodology, straight from our index. Free, Mondays.