Just Talk to It – The No-Bs Way of Agentic Engineering

Summary

I’ve been more quiet here lately as I’m knee-deep working on my latest project. Agentic engineering has become so good that it now writes pretty much 100% of my code. And yet I see so many folks trying to solve issues and generating these elaborated charades instead of getting sh*t done. This post partly is inspired by the conversations I had at last night’s Claude Code Anonymous in London and partly since it’s been an AI year since my last workflow update. Time for a check-in. All of the basic ideas still apply, so I won’t mention simple things like context management again. Read my Optimal AI Workflow post for a primer. Context & Tech-Stack I work by myself, current project is a ~300k LOC TypeScript React app, a Chrome extension, a cli, a client app in Tauri and a mobile app in Expo. I host on vercel, a PR delivers a new version of my website in ~2 minutes to test. Everything else (apps etc) is not automated. Harness & General Approach I’ve completely moved to codex cli as daily driver. I run between 3-8 in parallel in a 3x3 terminal grid, most of them in the same folder, some experiments go in separate folders. I experimented with worktrees, PRs but always revert back to this setup as it gets stuff done the fastest. My agents do git atomic commits themselves. In order to maintain a mostly clean commit history, I iterated a lot on my agent file. This makes git ops sharper so each agent commits exactly the files it edited. Yes, with claude you could do hooks and codex doesn’t support them yet, but models are incredibly clever and no hook will stop them if they are determined. I was being ridiculed in the past and called a slop-generator, good to see that running parallel agents slowly gets mainstream. Model Picker I build pretty much everything with gpt-5-codex on mid settings. It’s a great compromise of smart & speed, and dials thinking up/down automatically. I found over-thinking these settings to not yield meaningful results, and it’s nice not having to think ab...