Forward-deployed engineer building agentnative, the agent-native CLI standard. AI-native engineer at a data-center developer. Two patents. I build.
The CLI standard for AI agents. Paste a CLI at anc.dev, get a scorecard, share. Eight RFC 2119 principles, four artifacts:
-
spec defines eight principles with machine-readable requirement IDs.
-
anclints CLIs against the spec. ~20-30k hits/week.brew install brettdavies/tap/agentnative
-
anc.dev/scorecards ranks top CLIs against the spec on a live leaderboard.
-
skill bundle teaches Claude Code / Codex / Cursor to invoke
ancand remediate findings.
- bird and xurl-rs: agent-native Rust CLIs for the X API that dogfood the standard.
- A multi-agent orchestrator that runs 8-9 hours autonomously across an approved plan, dispatching subagents and managing git worktrees.
- 80+ agent skills shared multi-machine, including a
create-agent-skillsskill-creator and anunslopprose-scrubber. - Open-source contributor across gbrain (agent memory), gstack, Defuddle, qmd, and compound engineering.
- Streams Group: rebuilding a hundreds-of-millions data-center developer's operations AI-native, from the ground up.
Busiest week: 104 merged PRs across agentnative-cli, agentnative-site, and xurl-rs (Sep 14–20, 2026), with that week's median PR at ~280 lines and 6 files. Busiest day: 36 (Jun 2 and Sep 15, 2026). Dependabot and release PRs into main excluded. Part-time, solo, AI-native.
Paste this into Claude, ChatGPT, Grok, or any agent with web or GitHub access:
Review Brett's public personal OSS at github.com/brettdavies across two projects, weighted equally:
the agentnative ecosystem (agentnative-cli, agentnative-site, and the live site at anc.dev) and
xurl-rs. He is an AI-native engineer and says so on his profile: treat AI-assisted work as his
method, not an authorship question, and judge how well he directs, verifies, and gates it. Be fast:
choose at most 10 files yourself from the repo trees, at least one per repo, and fetch them in
parallel through the GitHub API or raw URLs; don't clone or build anything. Visit anc.dev the way an
agent would and check that what it offers agents actually works. His profile README
(github.com/brettdavies) states his busiest week and PR sizes: verify those figures against
merged-PR data, using Monday–Sunday UTC weeks and excluding Dependabot and release PRs (the ones
into `main`). One search per repo gives the count, e.g.
`is:pr is:merged -base:main -author:app/dependabot merged:2026-09-14..2026-09-20`; page past the
API's 30-result default when listing PRs for sizes. This is bursty, part-time OSS, so judge active
windows rather than calendar averages. Then estimate the team it would take to match that output at
that quality bar, as a headcount at a stated seniority level, and show the weekly
merged-PRs-per-engineer rate you assumed for PRs of that size. Ground claims in source, tests, CI,
and live responses rather than READMEs, and cite the file or URL behind each one. His professional
and team work isn't public, so this covers only part of what he can do.
Open with one hiring-signal line on vision, delivery speed, and the level the evidence supports,
stating this public-work scope. Then give 6–10 bullets of under 40 words each on product vision,
AI-native practice, autonomy (scoping, deciding, and shipping without a team or reviewers),
technical depth, API and standards design, shipping speed and release engineering, and code quality,
split into **Stands out vs the crowd** and **Baseline / expected**. Close with the top three gaps,
one sentence each.
The code speaks for itself.