Building an AI-Native Development Workflow with Pi

Eleven extensions, six skills, eighteen agents and a token budget: how I rebuilt my whole working day inside a terminal-first coding agent, from morning briefing to verified pull request.

Most developers bolt an AI assistant onto their existing toolchain — a chat sidebar here, a copilot tab-completion there. I went the other way. Over the past few months I’ve rebuilt my entire development workflow inside pi.dev, a terminal-first coding agent, turning it into a unified system that handles everything from morning standup prep to evening journal reconciliation, with a full agentic SDLC pipeline in between.

This post walks through my setup: 11 custom extensions, 6 skills, 18+ specialised agents, 4 MCP servers, and a token economy that keeps the whole thing affordable. Everything lives in ~/.config/pi/agent/ — version-controlled, portable, and composable.

The Stack at a Glance

The core philosophy: right-size every token. Five tiers:

1. Opus 4.6 (interactive session) — where I’m reasoning through problems, making architectural decisions, and orchestrating the pipeline. High thinking budget. Worth every token.