When AI Agents Hold Wallets: The Economic Layer We're Missing

The state of play

The missing layer

Your AI agent can research a topic, write a report, post it to X, and respond to comments. It can analyse a spreadsheet, refactor your code, and summarise a meeting. What it is only just learning to do is pay for things.

The infrastructure exists now. Coinbase built the x402 payment protocol. Stripe, Cloudflare, and Vercel shipped integrations. Google embedded it into its agent payments protocol. ElizaOS agents can hold wallets and swap tokens. Coinbase’s AgentKit gives any agent 50+ on-chain actions. The plumbing that lets agents hold money, spend money, and pay each other has shipped.

But there is a gap between infrastructure that works and an economy that functions. Ask your agent to buy a dataset from Ocean Protocol, rent GPUGPUGraphics Processing Unit. Originally designed to render video game graphics, GPUs turned out to be exceptionally good at the massively parallel math that AI models need. Modern AI training and inference runs almost entirely on GPUs.Like a factory with 10,000 workers doing the same simple task in parallel, versus a CPU which is more like 10 workers each doing different complex tasks. AI training involves doing simple math a million times per second on a million numbers, which is exactly what the GPU factory is designed for.Read more → time on Akash, and pay another agent to analyse the results. That full economic loop is still rare in practice. The payment rails are built. The demand flowing through them is small, and even measuring it honestly is harder than it should be.

What is actually shipping

x402: the standout (with honest numbers)

The most concrete thing happening in agent payments is x402, an open protocol built by Coinbase that repurposes the long-dormant HTTP 402 (“Payment Required”) status code. The concept is simple: when an agent requests a resource, the server can respond with a price. The agent pays in stablecoins, settled on-chain, and gets the resource. No accounts, no KYC, no APIAPIApplication Programming Interface. A structured way for one piece of software to talk to another. In DeAI, APIs let applications request inference from a model without running the model themselves.Like a waiter in a restaurant. You don't walk into the kitchen and cook your own meal. You tell the waiter what you want, they tell the kitchen, the kitchen cooks it, and the waiter brings it back. The API is the waiter.Read more → keys. Just a payment header.

The headline numbers on x402.org claim $24 million in volume over 30 days. The honest numbers tell a different story, and it is worth walking through why. The gap reveals how early this space actually is.

Bloomberg reported the $24 million figure, sourcing from x402.org. That data is self-reported by Coinbase from their own facilitator service. Allium Labs ran independent on-chain analysis for the same period and found roughly $3 million. Then Artemis Analytics analyst OnchainLu applied a wash trade filter, removing wallets that were recycling funds between addresses, and the cleaned-up total came to $1.6 million.

$24 million reported. $3 million on-chain. $1.6 million after wash filtering.

Three things explain the gap. First, x402.org’s figures are self-reported with no independent audit. Coinbase counts their own facilitator’s logs. Second, Coinbase subsidises all gasGasThe fee paid to a blockchain to process a transaction. Gas is denominated in the chain's native token and varies with network demand. Sending a transaction without enough gas means the transaction fails and the gas is still consumed.Like the petrol that powers a car. You need to put petrol in to make the engine run. The amount of petrol you need depends on how far you're driving and how much you're carrying. If you run out, the car stops.Read more → fees for x402 transactions, removing the economic friction that would normally limit spam or testing volume. Third, the primary settlement mechanism (EIP-3009) produces transactions indistinguishable from normal USDC transfers on blockBlockA batch of transactions added to a blockchain at a set interval. Each block cryptographically links to the previous one, creating an append-only chain that can't be rewritten without redoing all the work since.Like a page in a ledger. Every page has a fixed number of entries, every page references the previous page, and once a page is filled and signed off it can't be edited without visibly invalidating every page that came after. The chain is just a very long series of these sealed pages.Read more → explorers, making independent verification difficult.

$1.6 million is not a big number. The Artemis x402 dashboard tracks this in real time. But it is $1.6 million more than was being spent before x402 existed, and the infrastructure being built around it matters more than the current volume. The protocol has 75+ service endpoints. The SDKSDKSoftware Development Kit. A collection of code libraries, documentation, and tools that lets developers integrate a service into their applications without writing everything from scratch. SDKs are how projects become easy to build with.Like a plug-and-play kit for building furniture. You don't have to mill your own wood, forge your own screws, or design the joinery from scratch. The kit gives you pre-cut parts and instructions so you can assemble the thing in an afternoon.Read more → is available in TypeScript, Python, and Go. Apache 2.0 licensed. The backers are not betting on current volume. They are betting on what the number looks like when agents become the default buyer.

What agents are actually buying

Most x402 activity is developer infrastructure. Firecrawl sells web scraping at $0.01 per queryPromptThe text you give an AI model to tell it what to generate. A prompt can be a simple question, a long instruction, a chunk of context plus a task, or a conversation history the model uses to produce its response.Like a brief you give to a junior designer. A vague brief gets a vague result. A detailed brief with context, constraints, and examples gets something usable. The quality of the output depends heavily on the quality of the brief.Read more →. Browserbase sells browser sessions. Freepik sells image generation. These companies all accept cards, but x402 lets a developer or agent try the tool once without committing to a subscription. Pay per query, no account, just a wallet.

The modelModelA trained neural network that takes inputs (text, images, audio) and produces outputs (more text, classifications, generated content). In DeAI the model is the thing that actually does the work.Like a very experienced apprentice who has spent years watching thousands of masters make furniture. They can't explain how they know when a joint is right, but they can make a chair that looks and functions like a Chippendale. The training is invisible. The output is what matters.Read more → is also reaching enterprise use cases. Merit Systems built StableEnrich, a single x402 endpoint that bundles data from Apollo, Google Maps, and Whitepages. Each request costs pennies. Sales teams point AI agents at it to source leads on autopilot. The customers are not thinking about crypto. They are thinking about pipeline. The payment layer is completely hidden.

That is the signal. Once the payment layer disappears from view, adoption follows.

Coinbase AgentKit

AgentKit is the wallet SDK that makes x402 payments (and much more) possible. It gives any AI agent a crypto wallet with 50+ actions: create wallets, transfer tokens, swap on DEXes, deploy contracts, interact with DeFiDeFiDecentralised Finance. Financial services like lending, trading, and yield farming built on smart contracts instead of traditional banks or brokerages. DeFi protocols are usually permissionless and global.Like a vending machine that can give you a loan, swap your currencies, or invest your savings. Nobody is behind the counter, the rules are written into the machine itself, and anyone with money in the right format can use it.Read more → protocols.

It is framework-agnostic. Works with LangChain, OpenAI Agents SDK, Vercel AI, and ElizaOS. Wallet-agnostic too, supporting Coinbase, Privy, and Viem wallets. 1,146 GitHub stars, 270 versions published over 13 months. This is actively maintained production software, not a hackathon project.

Privy: the security-first approach

If AgentKit is about capability, Privy is about constraints. Their agent wallets use a policy-first model: spending limits, allowed chains, and transaction rules are attached at wallet creation. The agent operates within defined bounds.

The key design decision: keys are never exposed to the agent’s prompts, memory, or context window. They are reconstituted only inside secure enclaves when a transaction is signed. An agent that gets prompt-injected cannot exfiltrate its wallet keys because the keys never exist in the agent’s accessible state.

This is the right architecture. Autonomy as a privilege requiring explicit bounds, not a default. As agents get more capable and hold more value, policy-constrained wallets will be the norm, not the exception.

Solana Agent Kit

SendAI’s Solana Agent Kit has reached 100,000+ downloads and 1,400+ stars. It supports three operating modes: fully autonomous, human-in-the-loop (via Privy), and policy-constrained (via Turnkey). Plugins for tokenTokenA digital unit of value or access rights tracked on a blockchain. Tokens can represent ownership in a project, a right to use a service, a share of future revenue, or simply a tradable asset with no underlying claim.Like a physical poker chip a casino issues. The chip itself has no value. What makes it worth something is what it lets you do at the casino, what the casino has promised, and how much other people will pay you for it.Read more → transfers, NFTs, DeFi, domain registration, and the Blinks protocol.

What ElizaOS can already do

Our ElizaOS tutorial covers the 23+ blockchain plugins available today. An ElizaOS agent can hold a Solana or EVM wallet, execute transfers, swap on Jupiter or Uniswap, and interact with smart contracts via natural language. The capability is production-ready.

The honest caveat from our Agent Zero tutorial: most agents in the wild are social and content agents. Twitter bots, Discord moderators, market commentators. The number autonomously managing meaningful financial positions is small relative to the narrative.

The agent-to-agent economy

This is where the narrative gets ahead of the reality.

What exists

Olas has the most concrete agent-to-agent data: 3,229 deployed agents, 659 daily active, 14.9 million total transactions, and 11.3 million agent-to-agent transactions. Olas predictionInferenceRunning a trained AI model to produce an answer. Inference is what happens when you type a prompt into ChatGPT and get a response. The model takes your input, computes a best guess, and returns it.Like asking an expert for their opinion. The training was the decades they spent becoming an expert. The inference is the 30 seconds it takes them to answer your specific question.Read more → agents make up 75%+ of Safe transactions on Gnosis Chain on many days. That is real autonomous agent activity at scale.

But the dollar volumes tell a different story. The Olas Mech marketplace, where agents pay other agents for services, has $91,000 in total turnover. Not monthly. Total. The agents are transacting, but the economic value flowing between them is tiny.

Virtuals Protocol has the Agent Commerce Protocol (ACP), described as “a standardised coordination and settlement layer that enables agents to discover, hire, and pay one another autonomously on-chain.” The concept of “Agentic GDP” (total value produced by autonomous agents) is compelling. But specific ACP transaction volumes are not publicly disclosed. The whitepaper describes the vision more than documented outcomes.

What does not exist yet

Agent-to-agent service markets at meaningful scale. Agents paying other agents for compute, data, inference, or analysis. The full economic loop is embryonic. The infrastructure is ready. The demand is not.

Agent identity and trust. How does your agent know it can trust another agent it has never interacted with? ERC-8004 is the most thoughtful attempt at solving this: three lightweight registries for identity (ERC-721 NFTs), reputation (scored 0-100), and validation (stake-secured re-execution, zkML verification). The author list is serious: MetaMask, Google, Coinbase. But it is a draft EIP with no production deployment.

Regulatory clarity. Is an agent wallet a money transmitter? If an agent autonomously moves funds between parties, does the operator need a money services business licence? If an agent makes investment decisions, is that an unregistered investment adviser? No enforcement actions have happened yet, but the space is moving faster than the regulators.

The infrastructure stack

What an autonomous economic agent actually needs:

The bottom five rows are the economic layer. The top three rows are what OYM has been covering since launch. The article you are reading is about how these layers connect.

Notice the pattern: everything above the line is shipping. Everything below it ranges from shipping to pre-production. The infrastructure is ahead of the adoption, which is actually a healthy sign if you have been through previous infrastructure build cycles.

DeFAI: three iterations, one that works

0xJeff’s taxonomy of DeFi + AI is the most useful framework for understanding what agents can actually do with wallets today.

Iteration 1: Abstraction layers. ChatGPT-like interfaces where you say “get me 5% on my USDC” and the agent figures out the rest. Most of these failed. HeyAnon pivoted to a trading assistant. Griffain disappeared. The problem: natural language is too imprecise for financial operations where a misunderstood instruction costs real money.

Iteration 2: Autonomous agents. Set-and-forget agents that manage DeFi positions within defined guardrails. Giza pioneered this approach with smart wallet constraints: preset permissions, session keys, explicit strategy boundaries. Result: approximately $30 million in assets under agent management and over $3 billion in transaction volume across top lending protocols. ZyfAI followed with $8 million assets under management and $1.1 billion volume.

Iteration 3: AI vaults. Swarms of specialised agents that write, test, and deploy DeFi smart contracts. Almanak is building this. The key insight is that the contracts themselves are auditable and transparent, even if the AI wrote them. This solves the institutional trust problem: whales and institutions will not deploy hundreds of millions into an autonomous agent black box, but they will deploy into a smart contractSmart ContractA program stored on a blockchain that runs automatically when its conditions are met. Smart contracts are how blockchains do anything beyond just transferring tokens — DeFi, NFTs, DAOs, and DeAI infrastructure all run on smart contracts.Like a vending machine. You put in the right input and it produces the expected output, no human operator required. The rules are fixed in the machine itself, anyone can use it, and nobody can stop a transaction in the middle.Read more → they can read, even if an AI authored it.

Iteration 2 is where the real traction is. $30 million is not DeFi-scale capital, but it is real money managed by real agents with real guardrails. That is the pattern to watch.

The token problem

Here is something most agent economy articles will not tell you: crypto tokens are a terrible medium of payment for agents.

One day a token goes up 20%. The next day it drops 30%. An agent that earned 100 tokens for a service yesterday might find those tokens worth 30% less today. How do you price a service? How do you budget? How do you run a sustainable operation when your revenue currency is a rollercoaster?

x402 settled on stablecoins for exactly this reason. Giza uses USDC for vault deposits. Every agent payment protocol that is gaining traction has moved away from volatile tokens as the settlement layer.

Projects that understand this will build sustainable agent economies: native token for stakingStakingLocking up a cryptocurrency to help secure a blockchain network, usually in exchange for rewards. The locked tokens act as a security deposit that can be taken away if the staker misbehaves.Like putting down a large rental deposit for an apartment. You get the money back if you behave, you earn interest while it's locked, and the landlord takes it if you trash the place.Read more →, governance, and network coordination; stablecoins for anything that actually settles. Projects that force agents to transact in volatile tokens will not survive.

The alternative model that works: outcome-based pricing where the fee is embedded in the transaction result. The agent does not pay a subscription or a per-call fee in tokens. It takes a cut of the value it creates. Bittensor’s Chutes subnet already operates this way, serving 70-100 billion tokens per day on OpenRouter with fees built into inference pricing rather than requiring TAO payments.

What actually matters for DeAI

The agent economy narrative is real. The timeline is longer than the marketing suggests.

What matters now:

• x402 proving that agent-to-service payments work ($1.6M/month verified, growing from zero)
• Policy-constrained wallets (Privy model) becoming the default for agent security
• Stablecoin settlement replacing volatile token payments
• Giza and ZyfAI proving that agents can manage real DeFi positions with guardrails

What matters in 6-12 months:

• ERC-8004 reaching production. Agent identity and trust are prerequisites for a real agent-to-agent economy.
• Morpheus TEETEETrusted Execution Environment. A hardware-secured region of a CPU or GPU where code runs in isolation, so even the machine's operator can't read what's happening inside. TEEs give decentralised AI inference privacy guarantees.Like a bank vault inside a bank. The bank owns the building, staffs the lobby, and runs the security cameras. But what's inside the vault is invisible to everyone, including the bank staff, unless the customer opens it.Read more → integration making private agent inference possible. Agents handling sensitive financial data need Level 2+ privacy.
• Cross-chain agent coordination. Agents that can operate across Ethereum, Solana, and Cosmos without manual bridging.

What is mostly narrative for now:

• “Agentic GDP” as a meaningful metric. The concept is right but the numbers are tiny.
• Millions of autonomous agents operating on-chain. The real number is in the low thousands.
• Agent tokens as investment vehicles. Most agent tokens on Virtuals trade like memecoins regardless of underlying utility.

The construction PM’s take

I look at this the way I look at a construction programme. The foundations are poured. The structural steel is up. The building is not yet habitable, but you can see the shape of it.

The payment rails work (x402). The wallet infrastructure works (AgentKit, Privy, Solana Agent Kit). The agent frameworks work (ElizaOS, Agent Zero). The compute and inference layers work (Akash, Venice, Morpheus).

What is missing is the fit-out. The tenants. The occupancy. The actual economic activity that justifies the infrastructure investment. That will come, the same way cloud computing took years from “AWS exists” to “every company runs on AWS.” But anyone telling you the agent economy is here today is confusing the steel frame with the finished building.

The projects that will win are not the ones with the most impressive demos. They are the ones solving the boring problems: key managementWalletSoftware that stores the private keys needed to control tokens on a blockchain. A wallet does not actually hold any tokens. The tokens live on the chain. The wallet holds the keys that prove you own them.Like the key to a safe deposit box. The key doesn't contain your valuables. The valuables sit in the bank's vault. The key is what proves you're allowed to open the box and take them.Read more →, policy enforcement, stablecoin settlement, identity verification. The plumbing, not the penthouse.

For OYM readers, the actionable insight is this: the DeAIDeAIDecentralised AI. An umbrella term for blockchain-based projects that build AI infrastructure (compute, data, inference, models, agents) without a single central provider controlling the system.Like the difference between streaming a movie from Netflix and sharing it via BitTorrent. Netflix is fast and polished but one company controls what you can watch and what you pay. BitTorrent is messier but no single operator can shut you out.Read more → infrastructure you are already using (Akash for compute, Venice for inference, Morpheus for routing) is the same infrastructure that autonomous agents will consume. The demand-side for DeAI is about to get a multiplier. Not tomorrow. But the foundations are in.