228 lines
4.7 KiB
Markdown
228 lines
4.7 KiB
Markdown
|
# The Emerging Agent Internet Stack
|
||
|
|
|
||
|
|
As AI agents become capable of acting on behalf of people and
|
||
|
|
organizations, a new technology stack is emerging. These technologies
|
||
|
|
are largely complementary layers, not direct competitors. Together they
|
||
|
|
enable agents to discover information, access tools, transact
|
||
|
|
economically, and collaborate.
|
||
|
|
|
||
|
|
## Layer 1: Knowledge & Data (The Asset)
|
||
|
|
|
||
|
|
This is the actual value being exposed.
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
|
||
|
|
- PaddleNet industry data
|
||
|
|
- NetworkSIG relationship graphs
|
||
|
|
- Personal Identity Documents (PIDs)
|
||
|
|
- Public lands databases
|
||
|
|
- Company databases
|
||
|
|
- Knowledge graphs
|
||
|
|
- Search indexes
|
||
|
|
- Research repositories
|
||
|
|
|
||
|
|
These are the underlying assets that agents want to access.
|
||
|
|
|
||
|
|
## Layer 2: APIs (The Plumbing)
|
||
|
|
|
||
|
|
APIs are the traditional machine-to-machine interface.
|
||
|
|
|
||
|
|
They expose data and functions.
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
|
||
|
|
- GET /companies
|
||
|
|
- GET /customers
|
||
|
|
- POST /search
|
||
|
|
|
||
|
|
APIs require:
|
||
|
|
|
||
|
|
- Documentation
|
||
|
|
- Authentication
|
||
|
|
- Knowledge of endpoints
|
||
|
|
- Custom integrations
|
||
|
|
|
||
|
|
Think of APIs as the raw plumbing connecting software systems.
|
||
|
|
|
||
|
|
Application → API → Database
|
||
|
|
|
||
|
|
## Layer 3: MCP (The Agent Access Layer)
|
||
|
|
|
||
|
|
MCP (Model Context Protocol) sits above APIs.
|
||
|
|
|
||
|
|
It allows AI agents to discover and use tools in a standardized way.
|
||
|
|
|
||
|
|
Instead of calling a specific endpoint, the agent sees:
|
||
|
|
|
||
|
|
Tool: search_companies
|
||
|
|
|
||
|
|
Description: Search company database.
|
||
|
|
|
||
|
|
Parameters: - query - location - industry
|
||
|
|
|
||
|
|
MCP provides:
|
||
|
|
|
||
|
|
- Tool discovery
|
||
|
|
- Tool descriptions
|
||
|
|
- Parameter schemas
|
||
|
|
- Standardized access
|
||
|
|
|
||
|
|
Think of MCP as an API designed for AI agents rather than human
|
||
|
|
developers.
|
||
|
|
|
||
|
|
Many MCP servers are simply wrappers around existing APIs.
|
||
|
|
|
||
|
|
Agent → MCP → API → Database
|
||
|
|
|
||
|
|
## Layer 4: Identity & Trust
|
||
|
|
|
||
|
|
Agents need to know:
|
||
|
|
|
||
|
|
- Who they are
|
||
|
|
- Who they are interacting with
|
||
|
|
- Whether information can be trusted
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
|
||
|
|
- OAuth
|
||
|
|
- Passkeys
|
||
|
|
- OpenID
|
||
|
|
- DIDs (Decentralized Identity)
|
||
|
|
- Reputation systems
|
||
|
|
- Digital signatures
|
||
|
|
|
||
|
|
This area remains highly fragmented and is likely to be one of the major
|
||
|
|
battlegrounds in the coming years.
|
||
|
|
|
||
|
|
## Layer 5: Payments (MPP, x402)
|
||
|
|
|
||
|
|
### x402
|
||
|
|
|
||
|
|
Based on the HTTP 402 status code:
|
||
|
|
|
||
|
|
402 Payment Required
|
||
|
|
|
||
|
|
Flow:
|
||
|
|
|
||
|
|
Agent requests resource → Server responds: 402 Payment Required → Agent
|
||
|
|
pays → Access granted
|
||
|
|
|
||
|
|
x402 is focused on:
|
||
|
|
|
||
|
|
- Micropayments
|
||
|
|
- API payments
|
||
|
|
- Agent-to-service payments
|
||
|
|
|
||
|
|
### MPP (Machine Payments Protocol)
|
||
|
|
|
||
|
|
MPP expands the idea further.
|
||
|
|
|
||
|
|
Its goal is:
|
||
|
|
|
||
|
|
Allow machines and agents to transact economically.
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
|
||
|
|
- Pay for data
|
||
|
|
- Pay for searches
|
||
|
|
- Pay for introductions
|
||
|
|
- Pay for computation
|
||
|
|
- Pay for agent services
|
||
|
|
|
||
|
|
MPP often uses x402-style payment flows underneath.
|
||
|
|
|
||
|
|
Think of MPP as the business transaction layer for machines.
|
||
|
|
|
||
|
|
## Layer 6: Execution (Agents)
|
||
|
|
|
||
|
|
This is where actual work occurs.
|
||
|
|
|
||
|
|
Examples:
|
||
|
|
|
||
|
|
- ChatGPT
|
||
|
|
- Claude
|
||
|
|
- Codex
|
||
|
|
- Custom agents
|
||
|
|
- Agent workflows
|
||
|
|
- Agent orchestration systems
|
||
|
|
|
||
|
|
Agents:
|
||
|
|
|
||
|
|
- Search
|
||
|
|
- Analyze
|
||
|
|
- Plan
|
||
|
|
- Execute tasks
|
||
|
|
- Communicate with other agents
|
||
|
|
|
||
|
|
# Putting It Together
|
||
|
|
|
||
|
|
A future transaction might look like:
|
||
|
|
|
||
|
|
User → Agent → Discovers PaddleNet MCP → Requests industry report → MCP
|
||
|
|
exposes tool → Cost = \$0.25 → MPP/x402 payment → Knowledge Graph
|
||
|
|
queried → Results returned
|
||
|
|
|
||
|
|
Or visually:
|
||
|
|
|
||
|
|
Agent → MCP → Payment (MPP/x402) → API → Knowledge Graph → Results
|
||
|
|
|
||
|
|
# How This Relates to iNDX, PaddleNet, and NetworkSIG
|
||
|
|
|
||
|
|
The most valuable thing being built is not the MCP server or payment
|
||
|
|
protocol.
|
||
|
|
|
||
|
|
The valuable asset is:
|
||
|
|
|
||
|
|
- PIDs
|
||
|
|
- People Graphs
|
||
|
|
- Knowledge Graphs
|
||
|
|
- Industry Indexes
|
||
|
|
- Relationship Networks
|
||
|
|
- Opportunity Graphs
|
||
|
|
|
||
|
|
These are the data assets.
|
||
|
|
|
||
|
|
Over time they may be exposed through:
|
||
|
|
|
||
|
|
- PaddleNet MCP
|
||
|
|
- NetworkSIG MCP
|
||
|
|
- TerraSIG MCP
|
||
|
|
|
||
|
|
allowing outside agents to discover and query them.
|
||
|
|
|
||
|
|
Eventually some interactions may become paid:
|
||
|
|
|
||
|
|
- Find experts
|
||
|
|
- Find opportunities
|
||
|
|
- Make introductions
|
||
|
|
- Access premium intelligence
|
||
|
|
|
||
|
|
where MPP/x402 provides the economic layer.
|
||
|
|
|
||
|
|
# Simple Mental Model
|
||
|
|
|
||
|
|
Think of the stack as:
|
||
|
|
|
||
|
|
- Knowledge Graphs = The asset
|
||
|
|
- APIs = The plumbing
|
||
|
|
- MCP = Agent-friendly access
|
||
|
|
- Identity = Who is asking
|
||
|
|
- Trust = Can they be trusted
|
||
|
|
- MPP / x402 = How payment occurs
|
||
|
|
- Agents = Who is doing the work
|
||
|
|
|
||
|
|
Or even more simply:
|
||
|
|
|
||
|
|
Data → APIs → MCP → Agents
|
||
|
|
|
||
|
|
Payments → MPP/x402
|
||
|
|
|
||
|
|
Trust → Identity/Reputation
|
||
|
|
|
||
|
|
The long-term trend is toward an Agent Internet, where agents can
|
||
|
|
discover resources, access knowledge, transact economically, and
|
||
|
|
collaborate across organizational boundaries using standardized
|
||
|
|
protocols. iNDX, PaddleNet, and NetworkSIG fit naturally into the
|
||
|
|
knowledge, graph, discovery, and trust layers of that emerging stack.
|
||
|
|
|