// FIELD NOTES
Written from the leading edge.
What we're learning as we build with AI in production — honest, current, and free of hype. Much of it technical; all of it real.
My AI Orchestrator Is Dumb on Purpose
Most people assume the AI system running my backlog must be smart. It isn't, and I built it dumb on purpose. Detent moves real work from a backlog to a merged pull request without anyone babysitting it, but the orchestrator itself knows nothing about software. Every bit of intelligence lives in files I wrote, in plain text, for each project. In the video I open the actual files from my production projects — no mockups — and this post is the deep-dive version, with the real config excerpts inline. [youtube:QvED7RHTKkI]
Read →
Detent vs Hyperagent
Hyperagent and Detent sit close enough together that people are going to compare them, but they are not trying to be the same thing. Hyperagent gives you hosted agent teammates that can work across SaaS tools, channels, schedules, and artifacts. Detent gives you a self-hosted workflow runtime that turns issues, gates, review rules, isolated worktrees, and a merge train into the system that keeps work moving. One is hosted convenience with broad integration reach. The other is control, reproducibility, and discipline around the work contract. That makes this a useful comparison, not a fight.
Read →
The Malware Was Never in the Repo
Mozilla's 0DIN team published a proof-of-concept where a developer clones a normal-looking GitHub repo, asks Claude Code to get it running, and ends up handing an attacker a reverse shell. The malicious payload never appears in the repository. It arrives in a DNS TXT record, fetched by a setup script, suggested by an error message, followed by an agent that was only trying to be helpful. I've been saying for a while that coding agents are part of your supply chain now. This is what that looks like when someone weaponizes it, and the fix has a lot more to do with how you structure work than with which model you run.
Read →
Harness vs Orchestrator
Everyone is racing to build a smarter AI agent. OpenClaw, Hermes, Copilot, Cursor — they all want to hand you one agent that can do more. I went down that road too, and it never fixed my actual problem. There are two completely different things being called an "agent" right now, and mixing them up is the whole confusion. One is a harness. The other is an orchestrator. Only one of them stops you from babysitting. [youtube:bgXZuWZGDv4] Short on time? Watch the 90-second version .
Read →
Stop Managing Agents. Manage Work.
I've been orchestrating multiple Codex and Claude sessions by hand for about six months. Not vibe coding — strict, my skills, my workflows. And what I kept running into was the context switching: jumping project to project, trying to manage multiple boards at once. It's virtually impossible. The kicker is it was never an AI problem. I had tons of AI credits sitting there unused. The bottleneck was me. So I built Detent — a system, not an agent — that runs my board, my workflow, and my merge train. This post is the deep-dive version of the video. [youtube:SE_Noc3T_DY] Short on time? Watch the 90-second version .
Read →
I Challenged AI to Optimize My Go Code
I teach an advanced Go optimization workshop, and I've run it dozens of times. I know exactly what's wrong with the code and exactly how to fix it. But I wanted to know — can AI follow the same profiling discipline? Not guessing, but actually measuring, profiling, and proving every change with pprof and benchstat.
Read →
Chrome DevTools MCP Kept Conflicting — 5-Line Wrapper Fixed It
Chrome DevTools MCP is one of the best tools I've added to my AI coding workflow. It gives your coding agent a real browser — click buttons, fill forms, take screenshots, run full end-to-end tests. I run three or four projects doing their own e2e testing simultaneously while I'm still writing and reviewing code. But the shared Chrome profile creates conflicts when you do that, and the built-in fix ( --isolated ) kills the login credentials you need for real testing. A wrapper script solved everything. [youtube-short:dmeRmkMs7l4]
Read →
Code Is Cheap Now
OpenAI just posted a GitHub discussion about shutting down community pull requests on Codex. They haven't done it yet — but the reasoning they laid out says everything. In their words, "code itself is no longer the scarce resource." The company that built the AI coding tools is telling you that typing code isn't the hard part anymore. And I've been saying this for a while.
Read →
OpenClaw Nails 10 GitHub Issues at Once — One Prompt, Zero Babysitting
10 GitHub issues. One prompt. I sent a single message to OpenClaw and walked away. It accepted the GitHub invite, set up its own profile, worked every issue, pushed 10 PRs, addressed Codex review comments, monitored CI, and pinged me on Slack when everything was done. This is the difference between AI coding tools and AI agents.
Read →
Master PR Reviews with Claude's Address Review Skill
PR review comments piling up? One command handles the entire review cycle — fix code, push, watch CI, reply, resolve threads, and request re-reviews. The /address-review skill turns a tedious multi-step process into a single command.
Read →
Automating Your Dev Workflow with AI: From Issue to Green PR in Minutes
What if every pull request you pushed was already tested, security scanned, and code reviewed before you even created it? That's exactly what the upgraded /start-issue skill does in my open source gopher-ai repo.
Read →
How to Actually Stop Claude Code from Committing to Main
CLAUDE.md says "NEVER commit to main" but Claude ignores it anyway. Here's how to use hooks for system-level enforcement that Claude CAN'T bypass.
Read →
Keep Your PRs Green: Never Have a Failing Pull Request Again
I figured out a simple trick using gh pr checks --watch to ensure I never leave a PR in a failing state. By adding a few lines to my CLAUDE.md, I've automated the entire CI feedback loop.
Read →
Give Your Terminal AI Super Powers
I'm a longtime Vim and TMUX user who briefly switched to Cursor for its AI features. The one thing I missed when returning to the terminal? Command K - the ability to ask "what command does X?" and get an instant answer. Here's how I replicated that with two simple scripts.
Read →
How I Use Claude Skills to Quickly Work Through GitHub Issues
When working on GitHub issues, I do the same thing every time - check for duplicates, determine if it's a bug or feature, research existing PRs, and set up my environment. With Claude Code skills, I've automated all of this into a single command.
Read →
Claude Code's gopls Integration Was Eating 30GB of My RAM
After upgrading to a new Mac Studio with 128GB of RAM, I was still experiencing memory pressure issues. The culprit? Claude Code was launching a separate gopls instance for every session - and with 40+ sessions open, that adds up fast.
Read →Want this applied to your business?
Everything we write about, we also build and teach. That's the whole point.