The AIInterface
Bartley editions are built with AI assistants as first-class readers.
Readable by people or machines
Bartley editions are built for people who need knowledge they can trust, cite, and use. They are also built so AI agents can navigate that knowledge quickly and cleanly.
When an AI encounters a typical web site, it must parse its HTML directly, which is noisy and error-prone. When an assistant works with a Bartley edition, it can directly retrieve structure, articles, metadata, and perform semantic search across the book.
Bartley Editions provides three AI-first interfaces: Markdown mirrors, llms.txt files, and MCP servers.
Markdown mirrors
Markdown is a lightweight plain-text format that’s easy for both humans and machines to read and parse, making it ideal for preserving content in a structured yet accessible way. Every published page of a Bartley edition has a clean Markdown twin. The HTML page is for reading in a browser, while the Markdown mirror gives AI agents the same article without navigation, styling, or human-facing “furniture” in the way.
Here is an article from The Encyclopedia of Agentic Coding Patterns:
You find its markdown twin by simply appending .md to the URL:
llms.txt
Each edition publishes an llms.txt file, which is an emerging standard for a simple index that tells AI agents where the useful knowledge lives.
MCP Server
The Model Context Protocol (MCP) is an open standard that enables AI agents to connect to tools of all kinds in a structured way, including structured bodies of knowledge.
A published Bartley edition is also an MCP server that facilitates agents asking for corpus metadata, specific articles, related entries, and semantic or keyword search results.
For example, Claude Code (or any agentic harness) can connect directly to the Encyclopedia of Agentic Coding Patterns (EACP) MCP server:
claude mcp add --transport http eacp https://mcp.aipatternbook.com/mcpAfter that, the harness can query the EACP as a structured source instead of relying on general web search or pasted excerpts:
> What are some EACP articles related to evals?
• I'll search the EACP corpus for articles related to evals.
Called eacp
• Let me pull the relation graph around the core Eval pattern to round out the map.
Called eacp
• Here are the EACP articles most relevant to evals, grouped by how they relate.
Core
- Eval — the central pattern: a repeatable suite that measures how well an agentic workflow performs. Applies the discipline of testing to the agent itself rather than to its output. (Section: Agent Governance and Feedback)
- LLM-as-Judge — use one model to score another's output against a written rubric, so you can evaluate non-deterministic agent work at machine cost. (Correctness, Testing, and Evolution)
- Generator-Evaluator — split code creation and code critique into separate agents so neither role blinds the other. (Agentic Software Construction)
The mechanism evals plug into
- Verification Loop — run the work, check the result against an oracle, iterate until it passes. (Eval refines this.)
- Feedback Sensor — any post-action check that tells the agent what went wrong. (Evals are consumed by the agent through these.)
- Test — deterministic executable claims; evals are the non-deterministic analog. (Eval depends on it.)
Observability & governance around evals
- Agent Trace — the structured per-run record (spans for model calls, tools, sub-agents) that evals score against.
- AgentOps — operating/monitoring agents in production (complements Eval).
- Metric — quantified improvement/degradation signals (complements Eval).
- Runtime Governance and Permission Classifier — judging/ruling on agent actions at machine speed.
The two best starting points are Eval (the concept and discipline) and LLM-as-Judge (the workhorse technique for scoring non-deterministic output). Want me to pull the full text of either?
Why this matters
Together with human curators, the Bartley Engine creates coherent bodies of trustable knowledge. Editions remain a human-friendly, readable, citable works. The difference is that the knowledge is also available in forms that AI assistants can also navigate directly, and without friction.