Cloudflare Launches Flagship Feature Flags

Cloudflare is rolling out Flagship, a new native feature flag service built to handle the accelerating pace of AI-generated code. As AI agents increasingly contribute to and even autonomously deploy code, managing releases safely becomes paramount. Flagship aims to provide the necessary guardrails.

AI’s role in software development is rapidly expanding, from writing code snippets to shipping entire features. The next frontier involves AI agents autonomously managing the entire lifecycle, including deployment and iteration. This shift necessitates robust mechanisms to control the blast radius of new code, a role feature flags are uniquely positioned to fill.

"Flagship is now available in closed beta," according to Cloudflare.

The Edge Advantage for Feature Flags

Traditional feature flagging often involves network calls to external services, introducing latency. This becomes a bottleneck when your application, like Cloudflare Workers, runs at the edge, milliseconds from the user.

Hardcoding flags directly into Workers, a common workaround, quickly becomes unmanageable as the number of flags grows across teams, leading to a lack of central visibility and audit trails.

Cloudflare’s Flagship addresses this by running entirely on its global network. Flag configurations are synced to Workers KV and evaluated directly at the edge within Worker isolates. This eliminates external dependencies and latency, ensuring rapid flag evaluation.

OpenFeature Compatibility and Worker Integration

Flagship is built on OpenFeature, a CNCF open standard for feature flag evaluation. This approach allows developers to write their flag evaluation logic once and easily switch providers without extensive code refactoring, offering portability and avoiding vendor lock-in.

For Cloudflare Workers developers, Flagship offers a direct binding. This integrates flags directly into the Worker runtime, bypassing HTTP round-trips entirely.

The integration is straightforward, requiring minimal configuration in wrangler.jsonc. Developers can then access flag values using typed accessor methods like getBooleanValue() and getStringValue().

The service supports boolean, string, number, and JSON object flag types, enabling flexible use cases from UI variations to complex configuration management.

AI Agent Feature Flags and Safety

As AI agents gain autonomy, enabling them to ship code directly to production requires sophisticated safety mechanisms. Feature flags allow these agents to deploy new code paths in a controlled manner.

An AI agent can deploy a new feature behind a flag that is initially off. It can then enable the flag for a small test cohort, monitor performance, and ramp up the rollout or disable the flag if issues arise. This workflow, akin to what’s discussed in discussions around AI agent feature flags, keeps humans out of the loop for routine deployments while maintaining control.

This capability is crucial for the safe adoption of AI-generated code, whether it's from tools like Claude Code or future autonomous coding agents.

Flagship's architecture supports complex targeting rules, including nested logical conditions and percentage-based rollouts. This granular control is essential for testing and gradually releasing AI-driven features.

What's Next for Flagship

Flagship provides a full audit trail for all flag changes, ensuring transparency and accountability. A dashboard interface allows teams to manage flags without code deployments.

The service is currently in private beta. Cloudflare plans to share pricing details closer to general availability.

Developers can get started by requesting beta access, installing the SDK, or exploring the documentation and source code.