Series

AI Harness Engineering

7 posts+12 children
01

Coding Agents

Reflect on what I'm thinking and doing in this LLM era

12 parts
Cowork as PlannerClaude Code SetupClaude CodeClaude DesignCLAUDE.md, AGENTS.mdduyet/claude-pluginsInterview ModeLong-running and self-improving coding agentopencodeParallel Sub-Agents / TeamsPlan Modez_claude, mi_claude & or_claude
02

Claws

OpenClaw, NanoBot, PicoClaw, ZeroClaw, NullClaw, NemoClaw, ...

03

Cloudflare is all you need

Cloudflare's Agents SDK runs stateful TypeScript agents on Durable Objects, with Workers AI, AI Gateway, AI Search, Vectorize, Browser Run, Queues, Workflows, Email, Dynamic Workers, and Sandboxes around it. Plus Flue, the open agent harness from the Astro team Cloudflare acquired.

04

Letting Claude Code work overnight, 2026

A quick snapshot of where coding agents are in 2026: you can hand one a real, multi-part task, walk away, and come back to merged PRs, a live deploy, and a few issues it worked around on its own.

05

Autonomous Claude Code Migrate Nextjs to Tanstack Start

The second ultrawork run: I asked Claude Code to migrate chmonitor.dev from Next.js to TanStack Start and went to sleep. It planned the work, spawned parallel agents, and kept going for days.

06

Goal and Loop

You've probably heard of the agent loop by now. This is just a quick note on how I actually use it day to day — a `goal` and a `loop` — to get a lot more shipped.

07

Agent Sandbox on Kubernetes

I run Claude inside agent sandboxes on a home Kubernetes cluster. One consumes issue tickets into PRs, another triggers on Sentry webhooks to diagnose and patch runtime bugs. Here is the bigger loop, and where the new Kubernetes Agent Sandbox CRD fits.