Give AI a source it can answer from.

The source comes first

Each case study starts as one content entry. Its narrative, evidence, role, and outcomes live together. A structured field calledqaContext sits beside the prose with a short summary, checkable facts, and, when needed, a note about the limits of the claim. I can correct the source once instead of maintaining a second set of answers for an agent.

The human page and its Markdown twin are built from that entry. The twin is linked at the end of every published case study. Comparethe case study about Arcos Control Tower’s handoffwith its Markdown version. The format changes; the account does not.

Several paths, one account

An agent can start with the short site index, read the full generated document, or request a specific case study through the site’s read-only MCP server. Those paths use the same published content and renderers. The server returns records and source links; it does not invent an answer or make a hiring judgment.

Structured data in the pages gives crawlers another way to identify the work. For a person browsing normally, the case studies and guided paths remain readable with JavaScript turned off. The agent layer is another way into the same work, not a separate story about it.

A conversation has different needs

The on-page chat assistant is a separate, model-backed feature with its own authored fact corpus. It can answer a visitor’s question in conversation. The static files and MCP server above do something narrower: they expose source material for other agents to inspect. Keeping those jobs distinct lets a reader verify a case-study claim without trusting a chat response.

The same rule guides the build

I also work with an AI coding agent on this site. The repository holds the project rules, including the content voice, attribution, and requirement that guided pages still work without JavaScript. I make the product calls and review the resulting code and copy. The model helps me implement decisions; the written source and the shipped page are what I check.

The deeper account of the agent layer and the decision to extract it into a public template is inThe Content Is the API. The colophon covers the broader site stack.