ambient-code

ambient-code

Stop writing code. Start shepherding it.

License: MIT GitHub followers


The Vision

In AI-first development, engineers don't craft code line by line β€” they become shepherds of codebases, establishing guidelines and policies that optimize project goals, wielding entire agentic development teams as instruments.

We see future systems as a beehive of activity: agents replacing autoscalers, replacing algorithms, constantly regenerating in every direction β€” coordinating with each other, being as proactive as safely possible. The system is constantly evolving itself.


The Paradigm Shift

Traditional Code Era Ambient Code Era
Written once, maintained forever Generated fresh each time
Stored in repositories Created on-demand
Human crafted AI Generated with human expertise
Version controlled Specification Controlled & architected
Debugged line by line Validated through tests and data
Static artifacts Dynamic systems

The Code Shepherd

Developers evolve into Code Shepherds who:

  • 🎯 Establish Success Criteria - Define what good looks like through comprehensive specification and test development.
  • πŸ“‹ Set Generation Policies - Create guidelines that shape all code output
  • 🎼 Orchestrate Teams - Wield Agentic development teams as instruments toward project goals
  • ⚑ Optimize Systems - Continuously improve system efficiency and effectiveness

πŸ› οΈ The Pragmatic Implementation

Making Ambient Code Real

Today's technology stack enables ambient code concepts through proven components working together in the aggregate:

Component Role in Ambient Code
πŸ”΄ Test-Driven Development (TDD) Establishes success criteria that developers define
πŸ“‹ Application Specifications (SDD) Policies that guide infinite code generation
πŸ€– CodeGen LLM (agnostic!) The generation engine producing code on-demand
πŸ‘₯ Code Shepherds Humans orchestrating teams and validating systems
πŸ’Ύ Caching Patterns accelerating generation quality (tbd)

βœ… Variance Eliminated

  • For vibe coding or POCs, variance is acceptable.
  • Specifications constrain generation possibilities
  • Tests validate every generated artifact
  • Only conforming code enters production

βœ… Coordination Simplified

  • The limited structure created by spec-kit ensures no input gaps
  • Tests become universal contracts
  • Specifications ensure team compatibility
  • Proven patterns guide consistent output

βœ… Support Revolutionized

  • This is the mindset shift
  • Support the system behavior, not its implementation
  • Any passing code is valid code
  • Focus shifts from debugging to optimization

πŸš€ Ambient Code Projects

Platform & Ecosystem

Repository Description Language
πŸš€ platform Virtual team management and collaboration platform β€” ambient-code.ai Go
πŸ”Œ mcp MCP server for ACP (Ambient Code Platform) Python
🌐 browser-extension Chrome extension for managing ACP sessions
πŸ“± mobile ACP Mobile β€” React Native app for AI session management HTML

πŸ“– Platform User Guides β€” What is Ambient? Β· Quick Start Β· Core Concepts Β· Workflows Β· CLI Reference Β· Public API

Developer Tools

Repository Description Language
πŸ” agentready Repo Optimizer: Assess git repositories for AI-assisted development readiness Python
πŸ“‹ workflows Schema'd, versioned file to express the entirety of team SDLC preferences Shell
🧭 steering AI steering guidance generator β€” helps agents discover and use correct code abstractions Python
⚑ ambient-action GitHub Action that queries Langfuse for agent corrections and creates ACP improvement sessions Python
πŸ‘€ pull-reviews Pull request review tooling TypeScript
πŸ€– amber Ambient Code Organization Agent Python

Reference & Knowledge

Repository Description Language
πŸ“š reference Ambient Code Reference Repository β€” AI-assisted development best practices Python
πŸ—ΊοΈ review-roadmap Ambient Code Review Roadmap Python
🧩 session-config-reference Reference repo demonstrating every Claude Code session configuration surface
πŸ“‘ gps GPS β€” read-only MCP caching tier for org and engineering data Python
πŸ”— mcp-atlassian MCP server for Atlassian tools (Confluence, Jira) Python

Infrastructure

Repository Description Language
πŸ—οΈ opentofu Infrastructure as Code definitions HCL
πŸ”§ ops Operational scripts and tools for Ambient Code Platform