The Missing Layer for Autonomous Agents: Introducing the Autonomous Company Interface (ACI)

The Problem

An AI agent can call thousands of APIs. It can write code, query databases, and invoke language models. But there's one thing it cannot do reliably: Determine what an organization is, what it offers, and how it should be trusted - without scraping, inference, or bespoke integration.

When an agent reaches a company website, it sees pages designed primarily for humans - not a stable organizational contract. It can scrape text and make inferences, but it cannot reliably distinguish between a product, a pricing tier, a support contact, or a certification. It cannot reliably determine what identity claims, attestations, certifications, and supporting evidence the organization publishes, what APIs it exposes, or even which of its AI indexing tools are meant for autonomous consumption versus human reading.

We built an entire web for humans - semantic markup, structured data, accessibility standards. But we left autonomous agents to fend for themselves.

Why OpenAPI Isn't Enough

OpenAPI is excellent. It describes HTTP APIs with precision. But it describes interfaces, not organizations. An API definition tells you how to call an endpoint. It does not tell you:

• Which organization owns this API
• What business the organization is in
• Whether the organization is certified or regulated
• Where to find its other manifests
• Which of its offerings are available to autonomous agents
• What trust mechanisms it supports

An agent trying to decide whether to interact with a company needs organizational context first, API details second. That's the gap ACI fills.

What ACI Provides

ACI (Autonomous Company Interface) is an open specification that lets organizations describe themselves in a machine-readable format designed for autonomous agents. The core model is five manifest types:

In the reference implementation, each manifest is a linked JSON document. An agent discovers the first one - typically via /llms.txt - and follows links to the rest.

A Concrete Example

Consider a fictional company called NovaDynamics. At https://novadynamics.example/llms.txt, an agent finds:

# AI manifests
- [Identity Manifest](https://novadynamics.example/identity.json)
- [Capability Manifest](https://novadynamics.example/capabilities.json)
- [Knowledge Manifest](https://novadynamics.example/knowledge.json)
- [Trust Manifest](https://novadynamics.example/trust.json)
- [Agent Manifest](https://novadynamics.example/agents.json)

Fetching the identity manifest reveals:

{
  "manifest_version": "0.9.0",
  "last_updated": "2026-07-19T00:00:00Z",
  "publisher": "NovaDynamics Inc.",
  "jurisdiction": "US-DE",
  "website": "https://novadynamics.example",
  "description": "Autonomous infrastructure operations platform",
  "identifiers": [
    {
      "id": "org.novadynamics",
      "type": "domain",
      "value": "novadynamics.example"
    }
  ],
  "discovery": {
    "llms-txt": "https://novadynamics.example/llms.txt",
    "capability-manifest": "https://novadynamics.example/capabilities.json",
    "knowledge-manifest": "https://novadynamics.example/knowledge.json",
    "trust-manifest": "https://novadynamics.example/trust.json",
    "agent-manifest": "https://novadynamics.example/agents.json"
  }
}

The agent now has a stable identity anchor and machine-readable paths to NovaDynamics' capabilities, knowledge, trust assertions, and exposed agents. It can build an organizational profile from published contracts rather than site-specific scraping.

The Conformance Path

ACI defines three incremental conformance levels:

• Level 1 - Discovery: Identity, Capability, and an operational discovery chain
• Level 2 - Understanding: Knowledge, Trust, and resolved cross-references
• Level 3 - Interaction: Agent Manifest, an operational interaction method, and a validator score of at least 90

A pre-conformance starter - publishing only an identity manifest and /llms.txt - is a useful first step, but is not a formal conformance level. Level 1 takes about five minutes to implement. Create an identity.json, a capabilities.json, and an /llms.txt that links to them. Done.

How to Implement ACI Level 1 in Five Minutes

The fastest path - fork the template:

GitHub: narko4u/aci-pages-template

This GitHub Pages template includes everything you need: identity.json, capabilities.json, /llms.txt, and a landing page. Replace the placeholders with your organization's details, enable Pages, and you're live. No build tools required.

Or, do it manually:

• Create identity.json with your organization name, website, identifiers, and a short description
• Create capabilities.json listing your products or services
• Create /llms.txt pointing to both manifests
• Run the ACI validator to confirm:

  python3 validator/validate.py https://yourdomain.com
  

See a live example:

# Zero-dependency explorer - discover what Empire Labs exposes
python3 demo/aci-explorer.py empirelabs.com.au

What This Enables

Once ACI is adopted broadly, an autonomous agent can:

• Discover an ACI-enabled organization's published manifests without site-specific scraping
• Understand its declared products, services, capabilities, and domain vocabulary
• Evaluate identity claims, trust assertions, and linked evidence
• Discover declared agent endpoints and authentication requirements
• Supply structured organizational data to external policy and governance systems

This is the missing layer for autonomous commerce, supply chain automation, compliance checking, and agent-to-agent coordination at scale.

The Open Standard

ACI is published as an open draft specification under CC BY 4.0. The validator, schemas, and examples are MIT-licensed. Empire Labs serves as the initial steward. The governance documents define a path to transfer ACI to a neutral foundation or standards body after v1.0 and the published independence criteria are met.

The repository is at: https://github.com/narko4u/aci-spec

The specification, validator, and full examples are available now. ACI is currently in Draft v0.9. The path to v1.0 requires three independent implementations, community feedback, and a stable core. We are not declaring a finished standard - we are proposing one, and inviting the community to shape it. Empire Labs Pty Ltd is the initial steward of the ACI draft specification.

Next: Fork the ACI Pages Template, replace the placeholders, deploy it through GitHub Pages, validate the result, and submit your implementation to the Independent Implementation Tracker. Help us reach three independent implementations for ACI v1.0.