New review paper argues code is how AI agents think and act, not just what they produce

A new study asserts that the real barrier to intelligent agents lies not in the language model itself, but in the surrounding software layer. This layer includes tools, memory, tests, and permission limits, transforming the disconnected model into an effective agent. Deepseek is building a dedicated team called "Harness" in Beijing, reinforcing the idea that the combination of the model and the software layer forms the intelligent agent. The study is significant as it elucidates how intelligent agents operate, emphasizing the software aspects that dictate their behavior.