"François Chollet argues that software harnesses orchestrating neural-network calls can themselves be regarded as neuro-symbolic architectures."
In an August 2026 discussion, Chollet explicitly describes an inference-time software harness orchestrating neural-model calls as a neuro-symbolic architecture. He further characterizes the arrangement as symbolic software coordinating neural inference and tool execution, offering a broader interpretation of neuro-symbolic design than systems built around formal theorem provers.
View Source