| name | Fusion |
|---|---|
| last_updated | 2026-06-02 |
AI agent work is fragmented across many products, and each one locks you into its own models, workflows, and surfaces. A developer running agents can't drive them their way across machines without being captive to a single vendor's silo.
Fusion is the model- and surface-agnostic orchestration layer — neutral by design, with a plugin ecosystem so it adapts as models and workflows evolve, rather than betting the product on any one model or vendor.
Primary: The solo developer juggling 10 Claude Code terminals and 5 Codex terminals, bouncing between laptop and phone. They're hiring Fusion to keep every agent moving — across surfaces and machines — without dropping threads or losing track of what's running where.
- Concurrent agent sessions — agents actively driven at once per user; the core signal that orchestration is real, not idle.
- Active nodes — nodes running agents in a given week; proves the multi-node bet is being used, not just available.
- Ecosystem breadth — unique models + plugins active per user; the direct readout of the agnostic, pluggable thesis.
- Task completion rate — share of board tasks driven to done; outcome quality, not just activity.
- LOC shipped via Fusion — code actually shipped through Fusion-driven agents; a proxy for real work delivered.
Driving agents across machines and surfaces in a scalable, multi-node way.
Why it serves the approach: The surface- and machine-agnostic promise only holds if work scales out beyond a single box.
Meeting the developer wherever they are — dashboard/kanban, mobile, terminal.
Why it serves the approach: Surface-agnostic means no surface is the "real" one; every surface can drive and observe agents.
The plugin system, model swapping, and evolving workflows and structure — including missions, goals, and agent companies.
Why it serves the approach: This is where neutrality lives: the product evolves with the models instead of being locked to today's.
Unlocking multi-user collaboration through open, pluggable extension points rather than a hardcoded model.
Why it serves the approach: Same neutrality bet applied to collaboration — defined extension points keep the core open and let richer access, roles, and deployment models be built on top without forking the product.