Case study / 2025

AetherForge

Multi-agent collaborative IDE where humans and AI agents can edit the same codebase with CRDT-backed sync.

democode

CrewAI, Gemini orchestration, Automerge, React

AetherForge interface

Problem

Most AI coding interfaces assume one human and one assistant. AetherForge explored what changes when multiple people and specialized agents edit the same workspace together.

System Shape

  1. 01React client exposes a shared coding surface.
  2. 02Automerge keeps edits conflict-free across collaborators.
  3. 03Specialized agents operate on focused coding tasks.
  4. 04Gemini orchestration coordinates agent responsibilities and context.
  5. 05Human users retain visibility into changes instead of receiving opaque patches.

Highlights

  • -Used CRDTs to make collaboration a first-class system concern.
  • -Separated agent responsibilities instead of using one general assistant.
  • -Built around simultaneous editing, review, and iteration.

Lessons

  • -Agent collaboration is a distributed-systems problem wearing a UX hat.
  • -Conflict-free state matters before polish when edits are concurrent.
  • -Trust improves when users can see how agent changes are produced.

What I Would Do Next

Add stronger task provenance, per-agent diff review, and replayable collaboration history.