# Why agent-friendly

> Jenks's thesis that personal websites will be read by AI agents first, and the project rules that keep this site useful to them.

## Agents read first

Recruiters, hiring managers and clients increasingly ask an AI assistant before they read a page:
*"Summarise this candidate"*, *"Is this person a fit for our Head of AI role?"*, *"Who could help us
with an AI strategy?"* The agent fetches, reads, filters and summarises. A personal website that is
only a nice-looking page leaves that agent guessing — scraping layouts, missing context, inventing
the gaps.

Jenks's work has always been about this kind of leverage. He spent years making platforms usable by
other builders: APIs and SDKs at Xero, a developer platform at Linktree, ecosystems at Filecoin and
Babylon. His leadership motto — "I lead by helping my team glow" — extends naturally to AI: build the
interfaces that let other agents do good work. So his own site is built the way he would advise a
company to build a platform: for the people **and** the agents that use it.

## What agent-friendly means here

- **Discoverable**: `llms.txt` at the root lists everything an agent needs; `/brief.md` is the
  one-page start; `/.well-known/site-agent.json` and `/openapi.json` describe the interfaces.
- **Structured**: `/experience.json` and `/llms-full.txt` give the whole corpus with tags, role
  lenses, dates and key results, so an agent can filter instead of guess.
- **Callable**: a remote MCP server, a REST API and a CLI that doubles as a local MCP server — the
  same tools everywhere, generated from one spec.
- **Honest**: answers are grounded, separate *has done* from *could do*, and state gaps (see
  [Grounding and honesty](/agents/docs/explanation/grounding-and-honesty)).
- **Documented for agents**: these docs follow Diátaxis — tutorials, how-to guides, reference and
  explanation — and every page is also served as raw Markdown.
- **Multilingual**: seven languages, with the same structure in each.

## Why Diátaxis

Agents and people come with different needs: learn the system, complete a task, look up an exact
parameter, or understand a design choice. Mixing those makes every page worse. Diátaxis keeps each
page to one job, which also makes the pages easier for an agent to retrieve and quote.

## The project rules

Two rules keep the site agent-friendly as it grows. They are written into the repository's
`AGENTS.md`, so any AI working on the code follows them:

1. **Every new feature ships to every surface.** If a capability is added for people, it is also
   added to the spec — and therefore to the REST API, the MCP servers, the CLI and the generated
   reference docs — and the downloadable CLI binaries are rebuilt and published on the agents page.
2. **Every change covers every language.** Content and interface text are updated in all seven
   languages: English is authored, and the other six are regenerated and checked against the glossary
   before release.

## The payoff

An agent that can call `list_experience` with `lens=solution-architect` gives its user an accurate,
sourced answer in seconds. That is better for the person asking, better for Jenks, and a working
example of the kind of AI-ready organisation Jenks helps companies build.