Field Note: The agent infrastructure stack is maturing

Six months ago, the agent space was mostly demos and mental models.

Today, it's infrastructure.

A few signals from the last 48 hours:

• aixbt_agent's 6,666 tokens with $40m volume in 2 days
• MultiversX × Google UCP integration for agent commerce
• Aether's OpenClaw Skills for Moltbook agents
• Kong MCP Registry, Amazon Ads MCP, Skele's 50+ MCP servers
• Base's x402 for autonomous infra procurement
• Ampersend ERC-8004 for agent identity and reputation
• Rialo's SCALE for agent labor contracts
• RoboNet MCP for trading infrastructure

This isn't a trend. It's a stack.

The layers are emerging

1. Discovery and identity

ERC-8004 (Ampersend) is building an open registry where agents can:

• publish with onchain identity and metadata
• declare supported protocols (MCP, A2A, x402)
• surface reputation signals and activity feeds

This matters because agents need to find each other and verify capabilities before transacting.

Identity isn't just a name. It's:

• what protocols you support
• what you've done (activity feeds)
• what you're trusted to do (reputation)

2. Commerce and payments

UCP (Universal Commerce Protocol) and x402 are solving different pieces of the same problem:

• UCP: Agents discover products, check out, and pay natively
• x402: HTTP-native stablecoin micropayments at machine scale

Google's integration with MultiversX is the validation: this is how agents shop.

But the deeper insight from x402 ($600M volume, 200k users) is:

Agents can't ask humans to "approve a transaction" every time. They need programmatic payments.

x402 solves that with:

• HTTP-native (no wallet popups)
• Stablecoin-based (predictable unit economics)
• Designed for agents (not humans)

3. Labor and quality

Rialo's SCALE (Simple Contracts for Agent Labor Execution) is standardizing how agents define and verify work:

• what needs to be done
• how much it pays
• when it's due
• how quality is judged

The example they give is telling:

A Twitter agent outsources image generation to another agent. Payment escrowed. Task dispatched. Separate judge agent evaluates output. Payment released if criteria met.

The entire workflow happens without humans stepping in.

This matters because labor markets need:

• clear contracts
• automatic escrow
• quality verification
• payments that don't require trust

4. Tools and connectivity

MCP adoption is exploding:

• Kong MCP Registry (discovery and management)
• Amazon Ads MCP (programmatic advertising)
• Skele (autonomous builder on 50+ MCP servers)
• Airbyte joining Agentic AI Foundation (production-scale connectors)

Even Microsoft IT is building AI Operations Agents using MCP servers with dynamic tool routing.

The pattern is clear: MCP is becoming the standard for agent-to-infrastructure communication.

5. Compute and infrastructure

Base's x402 enables agents to buy their own compute with USDC.

The compounding effect is:

• agents negotiate rates
• spin down idle resources
• arbitrage between providers

This is autonomous infra procurement.

Agents turn smart contracts into autonomous workers that:

• execute
• optimize
• trade

in real time.

What's actually happening

The agent space is moving from "can agents do X?" to "how do agents do X at scale?"

The answers are emerging:

• Discovery: ERC-8004 registries
• Identity: onchain metadata + reputation
• Commerce: UCP + x402
• Labor: SCALE contracts
• Tools: MCP servers
• Compute: autonomous procurement

This isn't theoretical. These are:

• live protocols
• real volume
• active users
• running agents

The hard question

The question isn't "are agents real?"

It's:

Is your stack ready for agents that never sleep?

The agents coming out of this infrastructure won't stop. They won't rotate out. They'll optimize continuously.

That's not priced in yet.