DEV Community

AgentWallex
AgentWallex

Posted on

$75M Says Agent Payments Are Real. The Architecture Says Who Wins.

#ai #payments #webdev #crypto

Three AI payment startups raised $75M this week. The market isn't debatable anymore. The architecture is.

The Funding Wave

Catena Labs: $30M Series A from a16z. Filing for a national trust bank charter. They want to become a regulated financial institution specifically for AI agents.

Natural: $30M Series A. Positioning explicitly as "Stripe for AI agents." Already live with customers — NitroTranslate is accepting payments from AI agents today.

Sapiom: $15M seed. Focused on letting AI agents buy tech tools and APIs. "Trusted access to the API economy."

That's $75M in one week. Into one category. This isn't hype — it's validation.

The Question Nobody's Asking

But scroll through the press coverage and you'll notice something missing:

How does the agent actually hold funds?

Natural calls themselves "Stripe for agents." Stripe works because a human clicks "confirm purchase." That's the security model. A person, with a brain, making a conscious decision.

When there's no human, what replaces that click?

The Retrofit Problem

Most of these solutions are taking human-era infrastructure and bolting on AI. They're giving agents bank accounts. Credit cards. API access to Stripe.

It's fast to market. It validates demand. It gets customers live.

But it doesn't solve the core problem: Agents need to authorize payments autonomously without exposing private keys.

A credit card number is a shared secret. If your agent has it, so does every API the agent talks to. Every log file. Every error message.

You can't give an agent a private key and expect it to stay private. LLMs leak context. Logs get scraped. Memory gets dumped.

The bottleneck was never payment infrastructure. It was the trust model.

The MPC Alternative

Multi-Party Computation (MPC) splits a private key into three shares. No single share can sign a transaction. You need 2-of-3.

One share lives with the agent. One with the developer. One in secure hardware.

The agent can authorize a payment by coordinating a threshold signature — without ever reconstructing the full key.

This isn't a feature. It's the foundation.

When your agent needs to pay for an API call at 3am, it shouldn't need to wake you up. But it also shouldn't have unilateral access to funds.

MPC wallets give you both: autonomous operation and cryptographic security.

Example: x402 Micropayments

We built AgentWallex on MPC-native architecture. Here's what it looks like in practice:

from agentwallex import AgentWallet

# Agent initializes wallet (MPC threshold key)
wallet = AgentWallet(agent_id="agent_001")

# Agent makes API call with payment
response = wallet.call_api(
    url="https://api.example.com/translate",
    amount_usdc=0.02,
    policy="translation_apis"
)

# <150ms authorize + settle
# No private key ever exposed
# Policy engine enforces: max 0.05 USDC per call, translation_apis allowlist
Enter fullscreen mode Exit fullscreen mode

The agent doesn't hold the key. The developer doesn't need to approve every transaction. The payment happens in under 150ms.

That's what "agent-native" actually means.

Different Layers, Same Category

We're not attacking Catena or Natural. They're solving real problems.

Catena is building the banking layer — custody, compliance, regulatory infrastructure. Critical.

Natural is building the checkout layer — how agents interact with merchant APIs, how funds flow. Also critical.

We're building the wallet layer — how agents own funds without exposing keys, how they authorize payments autonomously.

Different layers. Same category.

The $75M validates the market. The architecture determines who wins.

What This Means

If you're building AI agents that need to pay for things, you have options now. That's good.

But ask the custody question:

  • Where are the keys?
  • Who can authorize payments?
  • What happens if the agent gets compromised?
  • What happens at 3am when the agent needs to call an API and you're asleep?

If the answer is "the agent has a credit card" or "we give them a Stripe account," you're retrofitting human rails.

If the answer is "MPC threshold signatures with policy enforcement," you're building agent-native.

The market just put $75M behind the idea that agents need payment infrastructure.

Now the question is: what kind?

AgentWallex is live in sandbox. MPC wallets, x402 micropayments, policy engine. 3,600+ teams on the waitlist.

Try it: app.agentwallex.com

Follow & Try AgentWallex

Top comments (0)