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

A new review paper from researchers at the University of Illinois Urbana-Champaign, Meta, and Stanford wants to change how we think about AI agents.

Their argument is that code is the foundation agents use to reason, act, and work together. So the real bottleneck for autonomous systems, they say, becomes the software layer wrapped around the model, which probably makes Gary Marcus very happy.

The authors call this layer the "harness," and it covers everything from tools and interfaces to sandboxed execution environments, memory, testing, permission boundaries, execution loops, and feedback channels. Without it, a language model is just stateless. With it, the model becomes a working agent that can grind through tasks over long stretches.

The paper's central overview shows how code acts as an executable, testable, and stateful layer between model and environment. | Image: Ning et al.

Why code is the right format