Introduction
Current agentic workflows typically involve decomposing user requests into sequences of tool calls with correctly resolved parameters, followed by processing results through reasoning traces in the language model's context window. The predominant method to enhance such reasoning is test-time scaling, training models to search over long chains of thought; however, this capability is entangled in model weights, is not verifiable step-by-step, and incurs high inference costs.
Forethought System
We present Forethought, a neurosymbolic reasoning system that treats reasoning as an explicit, verifiable program built from a library of symbolic and neural primitives, composed through a domain-specific language. The outcome is reasoning programs, which are concrete representations of the model's work and can be inspected and modified prior to deployment.
Evaluation and Results
Instantiated as a tool-calling execution kernel and evaluated across five benchmarks, Forethought improves base-model accuracy by about 30% relative and outperforms vanilla prompting, reinforcement learning scaffolds, and prompt-evolution methods, enabling small models to match or exceed frontier model capabilities.
In direct comparison, a non-reasoning model augmented with Forethought competes with a dedicated reasoning model while requiring roughly three orders of magnitude less post-training investment, remaining model-agnostic and auditable.
Blogger's Review: The emergence of Forethought opens new possibilities for transparency and verifiability in reasoning processes, effectively enhancing model performance through neurosymbolic programming, especially showcasing extraordinary capabilities in smaller models, which merits further exploration of its application potential in future research.