Dispute erupts as Jev-style AI decision models proliferate
TypeSafe's Jev release has prompted a wave of similar lightweight decision-model APIs from companies including Microsoft and OpenAI. Researcher Nandakishor Mukkunnoth claims he developed the underlying Jev architecture more than a year before TypeSafe's launch. These models return probabilities for direct use in code, avoiding the slower text-generation approach of conventional LLMs.
After TypeSafe introduced Jev, other firms—Microsoft and OpenAI among them—released comparable lightweight decision-model APIs, while open-source alternatives also appeared. Nandakishor Mukkunnoth of ConvAI Innovations says his earlier work, including a March 2025 arXiv paper on SalesRLAgent with published weights and data, anticipated the architecture; a September paper followed. He also launched Laya, an open-source, multi-language Jev-style system runnable on local hardware. Both efforts use "System One" terminology, echoing Daniel Kahneman's fast-thinking concept.
The spread of Jev-style models could change how developers embed AI decisions, offering faster, cheaper probability outputs for routine code logic. Businesses may lower inference costs, while researchers and small teams could face harder fights over credit and adoption. Open-source versions such as Laya may widen access for those running models locally. The dispute itself may shape trust, attribution norms, and how quickly lightweight decision APIs become standard tools.