Agentic One Pager¶
My take on the current state and future of agentic coding — condensed onto a single page.
Agentic One-Pager
CLI coding agents · workflows · tooling
Agentic means LLMs with a harness used for coding — aka coding agents. This page covers CLI agents only (OpenCode & Pi).
Agents
Getting Started
Start with a single agent in a terminal window. Give it detailed instructions and observe how it works — see what it does. Do not fire and forget at first.
Use permissions you're comfortable with. For me this means no git commit/push and read-only kubectl/helm commands (e.g. no kubectl delete ...). Note: If you want to be 100% sure you need sandboxing — you will never be 100% safe.
Set up notifications — this is what makes agents efficient. Switch to async: fire, forget, get notified. Works with OpenCode plugin or Pi plugin.
Agentic Development Environment
ADEOrchestrate agents in a fancy terminal
All use git worktrees — run multiple agents on one project in parallel. Switch between projects seamlessly.
Pick your orchestrator: