Cadenza — Multi-Agent AI Development Workspace (Personal)
Maximiliano Pantoja
Personal / experimental project.
Cadenza is an AI-powered development workspace that coordinates multiple coding agents inside a single workflow, built as a VS Code extension.
The idea
Most AI coding tools give you one assistant. Cadenza explores what happens when you treat different models as specialists with different responsibilities — one understanding requirements, another planning, another generating, another reviewing — with a lead agent coordinating between them and the developer staying in control.
It's an orchestration question more than a code-generation question: how do you assign work across agents, keep shared context coherent, and combine their outputs without the whole thing collapsing into noise.
How it works
Cadenza runs agents as separate processes, manages their tasks and context, and coordinates their outputs through a shared workspace. That makes it possible to combine different models — Codex, Claude Code, OpenCode — inside the same development session.
What I explored
Agent orchestration with a lead agent delegating to specialists
Curated prompts tuned per agent responsibility
A custom RAG layer for project context
Visualization of the relationships between agents
Token and cost management across parallel agents
Status
This is an ongoing personal experiment, not a shipped commercial product. I'm sharing it because the architecture questions it raises — context management, agent handoffs, cost control at scale — are the ones I find most interesting in applied AI right now.
Built with Node.js and TypeScript.
Like this project
Posted Sep 4, 2026
Personal experiment: a VS Code extension that orchestrates multiple AI coding agents as specialists in one workflow, with a lead agent, custom RAG and shared context management.
