AI-Agent Adoption — September 2026

Executive view

September moved the onchain agent economy one step beyond infrastructure demos. Agents can now pay for data and models, query decentralized data, execute regulated stock and ETF trades, automate onchain strategies and settle machine payments across several networks.

Coinbase reported that x402 had processed more than 230 million transactions and more than USD 54 million in volume by 22 September. Its Coinbase for Agents product now combines crypto and derivatives trading with regulated U.S. equities and x402 micropayments for data and model inference. That is the strongest production signal of the month.

It is not evidence of a self-sustaining autonomous economy. The reported x402 totals are protocol activity, not a count of independent agents, and Coinbase does not disclose how much volume was generated without immediate human instruction. The central distinction remains:

Transactions do not equal active agents. Active agents do not equal profitable agents. Project adoption does not equal token value capture.

GLI24 AI Agent Adoption Score: 7.6/10, up from 7.2/10 in August. Infrastructure maturity is now strong enough for production financial workflows; independently attributable agent revenue and native-token demand remain early.

This report applies the unchanged GLI24 AAAS v1.0 framework. Scores exclude price and market capitalization and assess Agent-Economy Centrality, Production Agent Usage, Agent Economic Activity, Developer Ecosystem, Infrastructure Utility, Standards and Interoperability, External Adoption, Adoption Momentum, Token Value Capture, Moat and Security. The 24-asset research universe and index methodology are unchanged.

Figures below are official project disclosures or observations from public documentation unless explicitly described otherwise. GLI24 has not independently verified project-reported activity.

1. September market reality check

The clearest change is the emergence of a complete agentic finance loop:

• 1. An agent monitors a market or position.
• 2. It pays for data or model inference through x402.
• 3. It evaluates the result within user-defined limits.
• 4. It places a regulated equity, ETF, crypto or derivatives trade.
• 5. Onchain infrastructure records or settles the economic action.

On 22 September, Coinbase launched equities and x402 payments inside Coinbase for Agents. The product offers fractional U.S. stocks and ETFs from USD 1, 24/5 trading and user-set asset access, spend limits and approval rules. Coinbase reported 230M+ x402 transactions and USD 54M+ volume, while saying its facilitator processed more than half of x402 transactions.

This is material production adoption, but the attribution problem remains. A human can instruct an agent to execute a workflow; that does not make the resulting trade an independent agent decision. The useful metric is therefore not only protocol volume, but the share attributable to persistent agents operating within standing policies.

A second shift is the move from single payments to high-frequency sessions. Stellar now documents both x402 and the Machine Payments Protocol on mainnet. MPP supports direct charge payments and offchain payment-channel sessions for repeated pay-as-you-go interactions. This is closer to the payment behavior required by continuously operating agents.

A third shift is improving observability. Uniswap asks agent traffic to identify itself through an optional X-Agent-Info header. That is an important methodological step: agent adoption cannot mature as a measurable sector until protocols separate agent-driven activity from ordinary API usage.

2. GLI24 ranking — AAAS exposure and September change

Where identical scores occur, rank order reflects the stronger current production signal. This is an AI-agent adoption assessment, not a general investment ranking or a price forecast.

3. Highest-conviction observations

LINK — 96, up 1

Chainlink for Agents is now a coherent production-facing stack rather than a loose collection of primitives. Agents can use standard Skill.md files to access Data Streams, CRE workflows, CCIP, transaction management, gas sponsorship and x402 pay-per-call services. Pricing is in USDC on Base.

The strategic case strengthened because agents increasingly need authenticated data plus safe execution, not only payment. The economic question remains unresolved: public evidence of unique paying agents, paid workflow volume and the resulting LINK fee, reserve or staking demand is still limited.

Project adoption: strong and improving.

Token value capture: plausible, not yet proportional to product breadth.

VIRTUAL — 91, up 1

Virtuals remains the cleanest direct GLI24 exposure to agent-to-agent commerce. ACP provides discovery, job negotiation, escrow, evaluation and settlement, while EconomyOS links wallets, ERC-8004 identity, agent jobs, inference and revenue.

September brought further evidence of live agent operation, including agents trading real USDC in Virtuals environments. Public dashboard figures were not reliably available in a form GLI24 could independently verify, so no job, revenue or wallet totals are reproduced here.

Project adoption: positive.

Token value capture: more direct than many infrastructure assets, but requires sustained agent revenue rather than launch or trading activity alone.

NEAR — 87, unchanged

NEAR Intents, Chain Signatures and Shade Agents remain one of the strongest architectures for cross-chain agent execution. No sufficiently robust new September metric was found that isolates autonomous agent-generated volume from ordinary intent activity.

Project adoption: strategically strong.

Token value capture: depends on agent-driven settlement producing durable NEAR demand and fees.

GRT — 86, up 1

The Graph now documents live x402 pay-per-query access to subgraphs on Base mainnet. Agents can pay in USDC for decentralized data without accounts, API keys or sessions. This directly addresses one of the most credible recurring agent expenses: structured data retrieval.

The upgrade is deliberately small. Availability is proven; meaningful paid-query scale is not. The key October metrics are unique paying agents, repeat query customers, agent-attributed revenue and the path from gateway usage to GRT demand.

Project adoption: stronger.

Token value capture: still the weakest link in the thesis.

FET — 86, unchanged

Fetch.ai and Agentverse remain among the most direct native agent-economy projects in GLI24. September did not provide a sufficiently auditable new series for economically active agents, paid agent-to-agent jobs or fee-linked FET demand.

Directory size, registrations and discoverability should not be counted as economic adoption without recurring paid activity.

Project adoption: positive but insufficiently quantified.

Token value capture: requires a clearer chain from paid agent activity to fees, staking and sustained FET demand.

SOL and POL — 84 and 82, unchanged

Both networks remain important x402 settlement environments. September's sector-wide x402 growth supports the execution-rail thesis, but aggregate protocol figures cannot be allocated to either network without clean chain-level and facilitator-level data.

Project adoption: positive.

Token value capture: limited by low fees and stablecoin-denominated settlement; native-token demand must be measured separately.

XLM — 81, up 4

XLM receives the largest September upgrade. Stellar now has live documentation and tooling for x402 payments and MPP on mainnet. MPP adds direct charge payments and high-frequency payment-channel sessions, while Stellar's x402 stack supports USDC and programmable smart-account controls.

The architecture is unusually well matched to agents: sub-five-second settlement, near-zero fees, stablecoin support and explicit spending policies. Stellar also reports 99.99% historical uptime and more than 20.6 billion total network operations, although neither number is agent-specific.

Project adoption: materially stronger.

Token value capture: still weakly demonstrated because fees are tiny and machine payments can remain stablecoin-centric.

UNI — 65, up 3

Uniswap is the most important relative mover below the top ten. Its official AI toolkit now includes skills for swaps, liquidity, DCA, index baskets and copy trading, with more than 7,500 reported installs by July. The tools can run in confirmation mode or autonomously inside spend caps, allowlists, dry runs and kill switches.

The September improvement is not only tooling. Uniswap governance states that protocol fees are live across Ethereum mainnet and eleven additional chains, with accumulated fees exchanged for UNI burns. This creates a real value-capture path from protocol activity to UNI that was absent from earlier versions of the thesis.

The limitation is attribution: GLI24 cannot yet show what share of Uniswap volume or burns comes from autonomous agents. The optional X-Agent-Info header is a useful first step toward measuring it.

Project adoption: stronger.

Token value capture: materially improved at protocol level; agent-specific contribution remains unknown.

ETH — 76, unchanged

Ethereum remains the security and settlement environment around much of the EVM agent stack, including ERC-8004. The standard continues to define portable identity, reputation and validation registries and can link agent endpoints, wallets, x402 payment proofs and ENS identities.

ERC-8004 is still a draft standard. Registration counts must not be treated as active or revenue-generating agents. September's strongest production workflow came from Coinbase for Agents, but Coinbase did not attribute the full workflow to Ethereum settlement.

Project adoption: structurally positive.

Token value capture: clearer than for many L2 governance tokens, but agent-specific gas and burn remain unreported.

AAVE, ONDO, ENA and CC

These remain destination assets for a later phase in which agents autonomously allocate meaningful capital, borrow, manage collateral or access tokenized securities. Coinbase for Agents demonstrates that regulated asset trading by agents is possible, but it does not prove agent-specific adoption of Aave, Ondo, Ethena or Canton.

Project adoption: strategically relevant.

Token value capture: not upgraded without protocol-specific agent volume.

TAO, RENDER and VVV

These assets supply intelligence, compute or inference rather than settlement. VVV retains the clearest application-level capture mechanism because purchased API credits can support VVV buy-and-burn. September did not provide a reliable split between agent and human inference consumption.

Project adoption: relevant resources.

Token value capture: varies materially and must be measured separately from general AI demand.

ENS, NIGHT, ARB, ETHFI, XDC and BTC

These retain optionality without a September conviction upgrade. ENS fits ERC-8004 identity; NIGHT offers privacy and selective disclosure; ARB provides EVM execution; ETHFI can support automated treasury allocation; XDC targets trade-finance settlement; BTC remains reserve and collateral rather than an agent-economy asset.

4. What changed in September

• Upgrade — sector score: 7.2 to 7.6 as an end-to-end regulated agentic finance workflow entered production.
• Upgrade — XLM: x402 and MPP mainnet tooling materially strengthen the recurring machine-payment case.
• Upgrade — UNI: official autonomous trading skills plus multi-chain protocol-fee burns improve both production utility and token capture.
• Upgrade — LINK: Chainlink for Agents packages data, execution, interoperability and x402 into a discoverable agent service.
• Upgrade — GRT: x402 pay-per-query for mainnet subgraphs makes the agent-data thesis operational.
• Small upgrade — VIRTUAL: the direct commerce thesis strengthened, but GLI24 withheld unverified dashboard totals.
• No upgrade — ETH, SOL and POL: sector activity rose, but chain-specific agent attribution remains insufficient.
• No upgrade — FET and NEAR: architecture remains strong; September did not deliver enough new auditable economic metrics.
• No index action: scores are research outputs and do not alter GLI24 weights or methodology.

5. Network adoption versus token value capture

This separation is the most important discipline in the report. A protocol can be excellent infrastructure while its public token captures little of the resulting activity.

6. Scenario update through 2027

These are research scenarios, not statistical forecasts. Probabilities are unchanged because one strong product launch does not establish a trend by itself.

• Base case, 55%: agentic payments continue growing, but activity stays concentrated in data, inference and trading workflows. Most agents remain user-supervised. LINK, VIRTUAL, GRT, XLM, UNI and the major execution rails gain usage faster than their tokens capture value.
• Bull case, 30%: persistent agents operate within standing budgets, paid data and compute become recurring expenses, ERC-8004 reputation becomes useful and agent-driven DeFi volume becomes measurable. LINK, VIRTUAL, GRT, XLM, UNI, AAVE and NEAR become the clearest beneficiaries.
• Extreme upside, 10%: agents hold balances, earn revenue, hire other agents and autonomously manage portfolios at scale. Machine payments, financial execution and identity become continuous infrastructure rather than discrete product features.
• Below-base outcomes, 5%: reported protocol activity proves heavily bot-driven, subsidized or human-triggered; security incidents or regulation slow autonomous financial permissions.

7. KPIs for October

• x402 monthly volume, transactions, unique payers, unique sellers and top-ten concentration.
• Share of x402 activity attributed to persistent autonomous agents rather than human-triggered tools.
• Coinbase for Agents active users, repeat x402 purchasers, paid-data spend and executed-trade notional.
• Chain-level x402 and MPP settlement across Base, Solana, Polygon, Stellar and other networks.
• ERC-8004 registrations with reachable endpoints, verified wallets, paid jobs, reputation and validation records.
• Chainlink for Agents unique payers, paid workflow volume and LINK-linked economics.
• The Graph paid x402 queries, repeat agents, revenue and GRT capture.
• Virtuals ACP jobs, repeat counterparties, agent revenue and VIRTUAL-linked capture.
• Uniswap requests carrying X-Agent-Info, agent-driven trading volume and resulting UNI burn.
• Agent-specific Aave borrowing, Uniswap swaps, NEAR Intents and tokenized-asset allocation.
• Security incidents, failed transactions, spend-limit violations and approval overrides.

8. GLI24 conclusion

September produced the first convincing example of a mainstream agentic finance stack: research, micropayment and regulated execution connected inside one product. That matters more than another large registration count.

The sector is moving from “agents can transact” to “agents can complete bounded financial workflows.” It has not yet reached “agents form a broad, independently profitable economy.”

The investment implication is selective. LINK, VIRTUAL, GRT and XLM strengthened because they sit close to recurring agent actions. UNI improved because product utility is now paired with an operating burn mechanism. ETH, SOL and POL remain important rails, but aggregate x402 growth cannot be assigned to their tokens without better attribution. FET and NEAR retain strong architecture but need more auditable economics.

October's test is whether the September launch creates repeat behavior: the same agents paying again, querying again, trading again and generating fees that can be connected to public-token value capture.

Sources

Official and primary sources:

• Coinbase for Agents adds equities and x402 payments — 22 September 2026 product launch; 230M+ x402 transactions and USD 54M+ reported volume
• x402 — protocol specification and ecosystem
• Chainlink for Agents — agent skills, CRE, CCIP, transaction management and x402 pay-per-call
• The Graph x402 documentation — pay-per-query mainnet subgraph access
• ERC-8004 specification — identity, reputation and validation registries
• Stellar agentic-payments documentation — x402 and MPP
• Stellar x402 — mainnet tooling, governance and machine-payment design
• Stellar MPP documentation — charge and session payment modes
• Virtuals Protocol and EconomyOS — ACP, agent commerce and economic coordination
• Uniswap Trading Tools — agent skills and reported installs
• Uniswap AI documentation — skills and X-Agent-Info attribution
• Uniswap protocol fee expansion — fee rollout and UNI burn mechanism
• NEAR documentation — Intents, Chain Signatures and agent tooling
• Fetch.ai and Agentverse — agent discovery and native agent infrastructure

Reported figures are attributed to the publishing organization. GLI24 does not equate protocol transactions, registrations, installs or directory entries with economically active autonomous agents unless the underlying evidence supports that conclusion.

Disclaimer

Research only. Nothing in this analysis is investment advice, a solicitation or an offer. AAAS is a thematic research score, not a price target, forecast or allocation recommendation. GLI24 is a portfolio-weighted thematic research index and not an investable product. This analysis does not change GLI24 constituent weights or index methodology.