John Detlefs

Agentic engineering, production reliability, and the discipline between them

Agentic Engineering · Production Reliability

I install the discipline that makes AI agents reliable in production.

Rules-as-memory, parallel agent orchestration through git worktrees, adversarial review before every push, and grounding for user-facing agents. Anchored in seven years of shipping customer-facing software where downtime costs real money — the dashboard and account surfaces behind $200M+ in annual revenue at GoDaddy, and cloud infrastructure for nine-figure consulting engagements at PwC (via Kunai).

Latest on the journal

What I actually do

April 16, 20264 min read

The first time I watched an agent catch one of its own bugs — not because I told it to, but because an adversarial review step I'd wired up flagged the diff before it could push — I thought: this is what code review was supposed to be. Not…

Read the full essay →

More posts and interactive demo

More from the blog

Who are you?

Browser fingerprint demo — same canvas hash as the full page, wired into analytics.

Recorded interactions (merged events for your fingerprint, cap 100)

Canvas fingerprint

Open the full demo

GitHub

GitHub commit activity over the last year

Recent public commit activity. github.com/jdetle

Guardian

Rust · Open source

A Rust daemon that surfaces host resource pressure to AI coding agents so they stop issuing requests the machine cannot serve — operational hygiene for teams mandating agents on a mixed laptop fleet. github.com/jdetle/guardian

Rules corpus

Agent discipline · This repository

Ninety-plus rules, each with an Origin section describing the class of defect it addresses — the memory layer a shipping agent needs and the most transferable artefact of the practice. .cursor/rules/

Currently at Kunai (PwC Network) as a Senior Cloud Developer. Previously Senior SDE at GoDaddy. Before that: full-stack across IoT (Meshify), crypto, and data tooling.

  • Agentic engineering / LLM tooling
  • Rust / systems programming
  • TypeScript / React / Next.js
  • AWS / cloud infrastructure
  • Observability / SRE
  • Experimentation / rollout design