# James Ross · Agent Operations Architect

> I measure what coding agents get done per dollar, and whether you can
> trust it, using cost, rework, review load, failure rates and spend. So far
> that's on my own production multi-agent workspace, and the next readings come
> from software teams. Agent Operations is the practice and Vector is the
> organisation. Australia. Contact: via https://jamesross.ai/#contact

## Pages

- [Home](https://jamesross.ai/): James's site, with capabilities, selected work and engagement models
- [Workspace map](https://jamesross.ai/workspace-map): the entire reference architecture on one page
- [Vector](https://vectorai.au/): the organisation, and the Agent Operations Review it offers
- [Learn](https://jamesross.ai/learn): two tracks over the 18 patterns. Seven practitioner modules (canonical home: the flagship repo's learn/) and a hands-on build on the starter template (docs/tutorial.md)

## Free tools (run in the browser)

- [Workspace Maturity Check](https://jamesross.ai/tools/maturity-check): 18-question practitioner assessment across six dimensions
- [Agent Readiness Assessment](https://jamesross.ai/tools/agent-readiness): 12-question organisation-level readiness readout
- [Redaction Pre-Flight](https://jamesross.ai/tools/redaction-preflight): client-side leak check for text about to be published
- [Context Carry-Cost Calculator](https://jamesross.ai/tools/context-cost): prices a context read by size, entry turn, and cache economics

## For agents

- MCP server: POST https://jamesross.ai/mcp (streamable HTTP). Tools: search_architecture, lookup_pattern, assess_workspace, price_context_read, check_redaction, and request_capability (tell it what you needed that isn't here)
- MCP resources expose the published patterns, learn modules and teardowns as text/markdown, one section per URI under architecture://
- Agent card: https://jamesross.ai/.well-known/agent.json
- Pattern taxonomy as data: https://jamesross.ai/patterns.json

## Markdown for agents

The pages above are served as HTML only. There are no per-page `.md` alternates and no `rel="alternate" type="text/markdown"` links, because the pages are hand-authored rather than generated and a second hand-maintained copy would drift from the first. Markdown reaches agents three other ways, all generated from source:

- the MCP resources under `architecture://` (the patterns, learn modules and teardowns, one section per URI)
- https://jimy-r.github.io/agent-workspace-architecture/llms-full.txt, the full text of the reference architecture in one file
- https://jamesross.ai/patterns.json, the pattern taxonomy as structured data

## Canonical technical material

- [agent-workspace-architecture](https://github.com/jimy-r/agent-workspace-architecture): the reference architecture, with 18 patterns, samples, evaluation method, teardowns and learn track. Agent-readable full text at https://jimy-r.github.io/agent-workspace-architecture/llms-full.txt
- [agent-workspace-starter](https://github.com/jimy-r/agent-workspace-starter): the architecture as a fork-ready template
- [awesome-agent-workspaces](https://github.com/jimy-r/awesome-agent-workspaces): curated ecosystem index
- [Questions](https://github.com/jimy-r/agent-workspace-architecture/discussions): the canonical Q&A home

## Writing

- [Substack](https://jimyr.substack.com/): essays for the people who own the problem, and a digest of new public work