WonderTwin AI – Local API twins for safe agentic development

Inspired by Simon Willison's recent post on dark software factories: https://simonwillison.net/2026/feb/7/software-factory/ , I started experimenting with the Digital Twin Universe concept for a fintech product I'm building. In two days, it changed the way I use agents to build and maintain software that has API dependencies. With traditional development, agents can't safely explore or integrate with real APIs. Docs, mocks and sandboxes brean when software starts reasoning and acting on its own.

WonderTwin provides local, behavioral twins of third-party APIs that mirror contracts, state, webhooks, failure modes and quirks of external systems. Agents (or humans) can test, develop and iterate safely, locally or in CI, without touching production. Without needing internet access. Just `wt install stripe@latest` and you have a compiled Go binary that fully simulates Stripe, compliant with their most recent SDK release. WonderTwins include an MCP server so agents can interact directly with each twin.

WonderTwin is open core, with a commercial offering for production teams. Latest twin versions are always and forever free, and you can also build your own twin with the included agent skill. The commercial offering offers historical versions, with chaos testing and other resiliency features coming soon.

If you're building or maintaining API-heavy systems, or experimenting with autonomous agents, I'd love feedback on whether this approach is useful and any ways it could be improved.