These AI systems that can perform user-oriented tasks are especially useful in optimizing enterprise workflows, making it a lot of people calling this the “year of AI Agents of AI Agents.” At the ServiceNow Annual Knowledge 2025 conference, the company collaborated with NVIDIA to introduce a new model to advance AI agents.
Apriel Nemotron 15b
On Tuesday, ServiceNow and Nvidia launched the Apriel Nemotron 15B, a new open source inference language model (LLM) built to achieve lower latency, lower inference costs, and agent AI. According to the release, the model was trained with domain-specific data from Nvidia Nemo, Nvidia Llama Nemotron post-training datasets, and ServiceNow.
Also: Nvidia’s 70+ projects at ICLR show that raw chip power is at the heart of AI acceleration
The biggest point of the model is to package advanced inference features in small sizes. This will ensure that the model runs as an NVIDIA NIM microservice on the NVIDIA GPU infrastructure, and still operates what enterprise-grade intelligence companies want.
The company shares that the Apriel Nemotron 15B shows promising results for the model category in benchmark testing, confirming that the model may be suitable for supporting agent AI workflows.
Also: Will synthetic data derail the momentum of generative AI or will it be a necessary breakthrough?
Inference capabilities are especially important when using agent AI, as AI performs end-user tasks in a variety of settings in these automated experiences. As you are performing tasks without human orientation, you need to do your own processing or inference to determine the best way to proceed.
Joint Data Flywheel Architecture
In addition to the models, the two companies also announced joint data flywheel architecture. This is a feedback loop that collects data from interactions to further refine the AI model. The architecture integrates the ServiceNow workflow data fabric according to the release and selects Nvidia Nemo microservices.
Also: Nvidia launches NEMO software tools to help businesses build custom AI agents
This collaborative architecture allows businesses to use enterprise workflow data to further refine their inference models, while also installing the guardrails they need to protect their customers, allowing data to be processed in a safe, timely manner, and providing the necessary controls. Ideally, this would be fed into the creation of highly personalized context-conscious AI agents, according to the company.
Get the top stories of the morning in your inbox every day Tech Today newsletter.