An Anthropic engineer stated in a video: "At Anthropic we don't write prompts anymore. We build loops and graphs." Over 30 minutes, she demonstrated how the Claude team built a system that lets the model prompt itself.

If you don't work in development, you can think of a prompt as the text a user sends to an AI model. What Anthropic is doing now is turning prompts into a structured workflow: a prompt becomes a node within the system, placed into the loop of an agent. The model generates the prompts it needs for the next step during execution, or the entire workflow is organized by a graph.

Manoj Siripalli shared this video on X: [Original Tweet](https://x.com/ManojSiri123/status/2090500140855660857). The evolutionary path presented is: Prompts → Agents → Loops → Graphs.

For general users, this shift can be understood like this: previously, you told AI what to do step by step; now, you give AI a goal, and it decides what to do next on its own. For developers, the more noteworthy takeaway is that prompts are being engineered into a node within the larger system.

While many people are still researching how to write better prompts, Anthropic has already hidden prompts inside the system itself. This shift says more about where the field is going than any prompt-writing trick ever could.