Cloudflare says its new Clef model means humans no longer need to be in the loop for AI agents

What changed

Cloudflare introduced two new AI decision models named Clef and Clef-flash. Both models are built on Qwen, an open-source foundation licensed under Apache 2.0. They are designed to enable AI agents to make structured decisions internally, without relying on text generation. Clef-flash stands out by delivering classification results in about 39 milliseconds, which is more than ten times faster than TypeSafe AI’s Jev decision model that it directly challenges.

Why builders should care

These models shift how AI agents handle decision-making by eliminating the need for human intervention in the loop. For developers and operators building automated systems, this promises faster, more reliable AI-driven decisions without the overhead of text-based reasoning or external guidance. Speed gains reduce latency in applications where quick, structured classification or decision-making is crucial, such as real-time moderation, automated workflows, or edge deployments.

The practical takeaway

Operators can integrate Clef models to accelerate decision tasks and reduce operational complexity by removing human checkpoints. The Apache 2.0 license also lets builders customize and deploy these models at scale without heavy vendor lock-in costs or restrictions. Faster classifications mean AI agents can handle higher volumes and more concurrent requests, which can lower infrastructure costs while improving user experience.

What to watch next

Watch how quickly Cloudflare pushes Clef into production environments and whether it gains adoption beyond experimental stages. Also track responses from competitors like TypeSafe AI and how their Jev model adapts to rising performance benchmarks. Finally, the evolving use of Qwen-based structured decision models will reveal if this approach becomes a new norm for autonomous AI agents in the wild.

AI Quick Briefs Editorial Desk