WorkOS Develops Horizon: An Internal Autonomous Code Factory That Orchestrates AI-Driven Development
Key Takeaways
- ▸WorkOS built Horizon, an event-driven autonomous code factory that automates the full orchestration layer of software development, not just individual coding tasks
- ▸The system is triggered by webhooks from existing tools and can handle complex features, multi-package implementations, and continuous improvement of its own reliability through use
- ▸WorkOS chose to build in-house because autonomous agents require secure sandboxes, company-specific context, and human-in-the-loop review—pointing to a broader industry insight that effective autonomous systems are company-specific
Summary
WorkOS has built Horizon, an internal autonomous code factory that extends beyond single-task AI coding assistants to create a continuous, event-driven system for full-stack software development. Unlike one-shot tools like Claude Code or Cursor, Horizon is triggered by webhooks from existing dev tools (Linear, Notion, Figma) and spawns long-running agent sessions that handle complex features across multiple packages, including testing, verification, and initial validation of acceptance criteria.
The system represents a fundamental shift in how AI can be applied to development workflows: by automating the orchestration layer itself—deciding what to work on next, setting up context, running verification, and preparing changes for production—Horizon enables a compounding feedback loop where each successful execution improves the platform. WorkOS built Horizon in-house rather than relying on external AI APIs, recognizing that effective autonomous systems require company-specific context, secure execution environments, first-class access to application signals (logs, errors, conversations), and human-in-the-loop review processes. The architecture was inspired by similar internal systems at Ramp (Inspect) and Stripe (Minions), but the team concluded that the most critical insight was not the pattern itself but the necessity of custom implementation tailored to each organization's infrastructure and culture.
- Horizon demonstrates the viability of continuous agent-driven development loops that free engineering teams from task assignment, enabling them to focus on requirements gathering and user acceptance
Editorial Opinion
Horizon signals an important inflection point in AI-assisted development: the transition from tools that accelerate individual developers to systems that can orchestrate entire development pipelines and improve themselves through use. The real competitive advantage is no longer in the LLM itself, but in how intelligently a company can integrate agents into their specific development workflows, infrastructure, and culture. As more teams build internal autonomous systems like this, we're likely to see a divergence between companies that succeed with AI-native development practices and those treating AI as just another IDE plugin.
