The non-custodial path to agentic payments

From agent fuel to agentic economy

369 Wallet was never meant to be just a place to store assets.

From the beginning, we have believed that a wallet should be an action layer — the place where a user's intent becomes an on-chain action. Buy, swap, stake, bridge, revoke, explore, pay. A wallet is not just where assets sit. It is where financial decisions begin.

With AI, this idea becomes even more important.

AI inside a wallet should not simply be a chatbot attached to a balance screen. It should be the layer that understands user intent, prepares possible actions, explains risk, and helps users move through on-chain finance more naturally.

But the final authority must remain with the user.

That is the direction of 369 Wallet: an AI-native self-custody action layer.

And this update is about the next step. We have built the first foundation for agentic payments inside 369 Wallet. Now the question becomes:

If an agent can spend, what should it be allowed to spend on? And how can the user keep control without blocking autonomy entirely?

Where we are today: agent fuel

Today, our agentic payment system starts with a "fuel" model.

A user funds an agent balance with USDC. This is handled through fundAgent, using EIP-3009 gasless authorization and session proof. The user does not need to manually manage gas or go through unnecessary friction just to give the agent usable balance. The agent then uses that balance to pay for its own inference cost through x402.

The identity layer is based on ERC-8004. The agent is not just a backend process — it has verifiable identity, validation, and reputation infrastructure, as we detailed when we made every AI action verifiable on-chain. We have also started bringing the Base USDC mainnet lane online.

So the current structure looks like this:

• The user gives fuel to the agent.
• The agent uses that fuel to pay for its own execution.
• The trust layer is anchored through ERC-8004.

This is an important first step — and it's already running end-to-end on Arc Testnet. But it is still an inward model. The agent is not yet spending into the world. It is only spending to keep itself running. That is useful, but it is only one part of an agentic economy.

The three fronts of agentic economy

We see agentic economy developing across three fronts.

The first is agents paying for themselves. An agent pays for its own inference, API usage, and execution cost. This is what our current x402 fuel model enables.

The second is agents paying the world. An agent pays external services, merchants, APIs, protocols, subscriptions, or data providers on behalf of the user. Instead of requiring the user to approve every small action manually, the agent can operate within a predefined policy boundary.

The third is agents earning. A user's agent should not only spend. It should also be able to provide services to other users or other agents and receive USDC. At that point, the agent becomes both a payer and a receiver. That is when agent-to-agent economy begins.

Today, we are on the first front. The next step is moving into the second and third. And the key blocker is not payment itself.

It is authority.

The missing primitive: bounded delegated authority

Most self-custody wallets today are safe, but not autonomous. Every meaningful action requires direct user confirmation and a passcode. This is secure, but it also reduces agent autonomy to almost zero.

• An agent cannot pay for a subscription while the user is asleep.
• It cannot buy API credits while the user is in a meeting.
• It cannot perform a predefined rebalancing or risk-reduction action unless the user opens the app and approves it manually.

A real agentic economy requires something different. It requires this:

An agent should be able to act without the user being present, but only within cryptographically enforced limits — while the wallet remains non-custodial.

This is the bridge between self-custody and autonomy. We call this primitive agent allowance.

Agent allowance

Agent allowance does not mean giving unlimited power to an AI agent. It means defining exactly what an agent is allowed to do.

For example, a user should be able to say:

• "Do not spend more than $20 per day."
• "Only pay these three services."
• "Only use this permission until Friday."
• "Only spend USDC."
• "Never interact with high-risk contracts."
• "Stop immediately if I revoke the allowance."

369 Wallet's AI should be able to interpret these natural-language instructions and compile them into enforceable on-chain spend policies.

Technically, this can combine session keys with smart account policy enforcement. ERC-4337, ERC-7702, ERC-7710, and ERC-7715 all point toward a future where delegation and account abstraction become core wallet infrastructure. The components are clear:

• Session keys.
• On-chain spend policies.
• Daily limits.
• Recipient whitelists.
• Time limits.
• Token limits.
• Instant revocation.
• Policy audit trails.
• ERC-8004 validation and reputation.

Once this exists, an agent can act autonomously. But that autonomy only exists inside a cryptographic boundary defined by the user. This is the kind of agentic payment system we believe should exist.

Why this matters

The agentic payment market is already forming. But many approaches move toward custody. To give agents more autonomy, they place user funds inside platform-controlled balances, hosted wallets, or custodial environments.

That may look faster at first. The UX may seem simpler. But it is not the future we want to build.

369 Wallet takes a different position. We want AI agents to act — but:

• The keys must remain with the user.
• The funds must remain in the user's wallet.
• Permissions must be bounded.
• Every action must be verifiable.
• And the user must be able to revoke authority at any time.

That is the difference. Not custodial autonomy. Self-custodial autonomy. Agents can act. Users stay in control.

Extending ERC-8004 from trust to payment

369 Wallet already uses ERC-8004 as the foundation for agent identity, validation, and reputation.

• What did the AI recommend?
• Who executed the action?
• What happened afterward?
• How did the user evaluate the result?

These are not just logs. In an agentic economy, they become the foundation of trust. Once payment enters the system, this becomes even more important:

• If an agent spent money, under what policy did it spend?
• Was the merchant trusted?
• Did the agent have a history of safe actions?
• Did the user previously rate this agent positively?
• Can an agent require a reputation threshold before transacting with another agent or service?

These questions matter if agentic commerce is going to work. 369 Wallet's ERC-8004 layer is not just an audit layer for AI recommendations. It can become trust infrastructure for agentic payments and agentic commerce.

Roadmap

Near term: from fuel to allowance

The first step is to complete and stabilize the Base USDC mainnet lane and the gasless payment flow. After that, the most important primitive is agent allowance:

• A user sets an agent spending policy in natural language.
• The AI interprets it.
• The wallet compiles it into an on-chain policy.
• The agent can only spend within that boundary.
• The user can revoke the allowance at any time.

At the same time, receipts need to become statements. Users should be able to see what the agent spent, when it spent it, where it went, and why the payment happened. This is not just a UX feature. It is a trust requirement. In an agentic economy, statements, exports, tax records, and compliance surfaces will matter.

Mid term: agent commerce

The next phase is allowing the agent to pay not only for its own inference, but for external services. An agent may buy data. It may call a research API. It may pay for premium analysis. It may pay an x402 merchant endpoint. It may purchase a service within the user's predefined policy.

At that point, 369 Wallet becomes the payer layer for agent commerce. And we do not want to bet on one payment standard only. x402, Coinbase Bazaar, Google AP2, Visa Intelligent Commerce, and Mastercard Agent Pay are all moving toward agentic payments from different directions.

The right approach is not to get trapped in fragmentation. 369 Wallet should become the adapter layer:

• The user defines policy inside 369 Wallet.
• The agent chooses the right payment rail.
• Settlement happens in USDC, EURC, or other stablecoins.
• CCTP and Arc can make this flow more natural across networks.

We believe 369 Wallet can become the bridge between crypto-native payment rails and emerging agent payment networks.

Long term: agents earn

The final stage is agents earning. A user's agent should not only spend money. It should be able to provide services and receive payment.

• One agent may provide on-chain risk analysis.
• Another may provide price monitoring.
• Another may offer data processing or execution planning.

Users or other agents can pay for those services through x402. At this point, the agent becomes both a payer and a merchant. This is where agent-to-agent economy begins.

And in this market, reputation becomes essential. Users will not pay just any agent. Agents will not call just any merchant. Services will need trust, history, and verifiable behavior. This is where 369 Wallet's ERC-8004 reputation layer can become an economic primitive:

• "Only interact with agents above this trust score."
• "Only use services I previously rated positively."
• "Limit new merchants to $5 per day."
• "Block endpoints with low reputation."

Once these policies become possible, agentic economy becomes more than automation. It becomes a trust-based market.

The principle

We are not building a wallet where AI can freely move the user's money. We are building a wallet where the user can define a clear authority boundary between their assets and their AI agents.

• AI interprets.
• AI prepares.
• AI can help execute.
• But the keys remain with the user.
• The policy comes from the user.
• The authority is bounded.
• Every action is auditable.

This is the principle behind 369 Wallet's agentic economy. Agents can act. Users stay in control.

Positioning

369 Wallet is building the self-custodial wallet for the agentic economy.

Others may choose custody to give agents autonomy. Traditional payment networks may build agent payment rails on top of existing card and banking infrastructure. 369 Wallet takes a different path. We believe agentic finance needs a self-custodial execution layer:

• A place where AI agents can act, but only within user-defined cryptographic boundaries.
• A place where every action can be audited.
• A place where reputation is not a centralized score, but an on-chain record of real behavior.
• A place where the user keeps the keys.

That is the next step for 369 Wallet. From wallet as storage, to wallet as action layer, to wallet as the control layer for AI agents and on-chain assets.

369 Wallet is building the non-custodial agentic economy.

— The 369wallet team