Agentic Engineering

Engineering with Agents

I don't use AI as a fancy autocomplete. I've built a structured system that turns Claude Code into a disciplined engineering teammate — planned, reviewed, and accountable at every step.

Spec-first

Every change starts as a task and a plan I approve before a single line of code is written.

Custom skills

A skill system encodes my conventions — Git flow, architecture, review lenses — so the agent works the way I do.

Parallel orchestration

Independent tickets fan out to multiple agents, each in its own isolated git worktree, then merge wave by wave.

Feedback loop

Lessons from each task are logged back into the skills. The system compounds — the agent gets sharper at my conventions over time.

Guardrails by default

Preview before action. Never push without approval. The agent accelerates the work; I stay the engineer making the calls.

Build with agents

Hiring for AI-native engineering, or want this workflow on your team? Let's talk.

Get in touch
Case Study

Claude × Cloudflare

What happens when agentic delivery meets an edge-native platform — a reference architecture for a resilient, multi-channel messaging system, built with Claude Code.

Multi-Channel Messaging Platform

A reference architecture for resilient, multi-channel message delivery on Cloudflare's edge — designed and built with Claude Code.

Cloudflare WorkersHonoDrizzle / D1Durable ObjectsQueuesR2Claude Code

The challenge

Reliable, multi-channel messaging — SMS, email, postal mail, and print. Every message has to be delivered, retried on failure, and tracked end to end.

The architecture

Nano-services

Composable Cloudflare Workers (Hono) — an API, message orchestration, dispatch, per-channel senders, and templating, each a small, independently deployable service.

Queues + dead-letter

Per-channel queues (SMS, email, postal, ack), each with a dead-letter queue and a monitor worker for resilient, retryable async delivery.

Durable Objects

Stateful per-message coordination at the edge, keeping delivery state consistent across the whole pipeline.

Edge data

D1 + Drizzle for relational data, KV for fast key-value storage, R2 for artifacts — all running at the edge.

Serverless cloud printer

A serverless cloud printer — staff add a printer and print from any app, and documents are converted and routed straight into the messaging pipeline.

Built with Claude Code

Designed and delivered through the agentic workflow above — planned, reviewed, and shipped with Claude Code.

Cloudflare's edge primitives map cleanly onto a resilient, scalable messaging system — and Claude Code built it. Claude + Cloudflare is a fast, dependable way to build resilient, multi-channel messaging at the edge.