DeepSeek is having its second DeepSeek moment It's opening up the layer many thought would be the next moat - the harness. Their DeepSeek Harness is the layer that turns a model into an agent: tools, memory, execution loop, sandbox, storage, scheduling, interface. (DeepSeek has released its version of that entire stack under an MIT license) The whole thing is built around one idea: "Everything is a Plugin." Even the model provider. So you don’t have to get the model, tools, and runtime from the same company. ▪️ But here’s the most interesting thing about Harness - the agent can build a missing piece of itself while it’s running. If it needs a tool that doesn’t exist, it can create a temporary plugin. What’s starting to appear is a loop: need a capability → generate the tool → use it → remove it. No retraining or new version. This self-evolving loop isn’t complete yet. But suddenly, you can see the architecture for it. So, R1 challenged the idea that frontier models would stay concentrated in a few labs. DeepSeek Harness is now putting the same pressure on the layer above them. And the next moat suddenly looks a lot more modular and a lot more open.