manifesto

Lights-off software factories.

Software simulations that complete the AI dev loop.

Lights-off software factory — Max and Dot beside a LIGHTS switch as code crates move on a conveyor

the problem

The problem

Software is being written by AI agents now. Every external service those agents touch — Stripe, Plaid, Shopify, every commercial API — breaks the dev loop at agent speed. Sandboxes were built for humans running occasional manual tests, not swarms of agents iterating hundreds of times against production-shaped services. Mocks lie. Production isn't safe.

the category

The category

The factory model has an answer. When every external service a software factory touches is twinned — local, behavioral, production-fidelity — the factory runs lights-off. Autonomous. At machine speed. Without humans patching every integration.

the solution

The solution

WonderTwins are local behavioral simulations of external services. Composable across vendors, drift-aligned with production, MCP-discoverable by every Claude Code, Cursor, and Windsurf agent. Your agents develop against them locally — before code ships, before sandboxes get bashed, before any real bank account or customer record is touched.

the mechanism

The mechanism

Agent-Led Expansion (ALX) makes the factory inevitable. Agent workflows surface what needs twinning. The registry auto-completes inside the account. At full maturity, customers grant standing authorization: anytime an agent finds a system that can be twinned, twin it.

the precedent

The precedent

LocalStack built a $25M-funded, enterprise-validated business on AWS-only coverage. We're doing the same for the commercial API layer.

Building agents that hit external services?

Want to make your API a first-class WonderTwin?