Announcing Data Agents: Apply Agents to Your Data With One Click

Introduction

The best AI product teams combine agility and adaptability, particularly in competitive and rapidly evolving domains like Physical AI. When a breakthrough VLM like NVIDIA Cosmos Reason v2 is released, AI product leaders need to start trying it out and find ways they can build on top of it quickly.

At Encord, the unified layer for AI development, we’ve worked alongside AI leaders across all verticals and company stages. From our work with Pickle Robot to Toyota, we’ve seen the persistent need for faster, production-ready data preparation processes like pre-labeling using foundation models, automated quality checks, and custom task routing and orchestration. 

Today, we’re excited to announce a revamped library of Data Agents, a set of agents that helps extend the Encord platform programmatically and eliminates the need for custom scripting or integrations.

Cosmos Reason v2 Agent Setup within Encord

Agent Spotlight: Integrating NVIDIA’s Cosmos Reason v2

We’re particularly excited to announce that we’re debuting this capability with instant deployment for Cosmos Reason V2, the recently announced VLM from NVIDIA. 

In the past, integrating humans and SOTA models into data workflows required custom development and diverted Engineering resources away from other priorities. Setting up VLM video captioning or annotator routing for a new model would have required custom code, slowing progress toward production-ready AI. 

With Encord’s new Data Agents, we’re embedding deployment best practices that we’ve gathered from across hundreds of successful customer engagements into point-and-click options for your AI team, making it easy to experiment with and adopt new models. 

Imagine a robotics startup that wants to use Cosmos Reason v2 for pre-labeling. They can now select the model from Encord’s in-app library and configure parameters for classification, captioning, or VLA. Below, we show how annotators within Encord’s platform can work directly on top of videos using Cosmos Reason for VLA. 

Our early engagements show that adopting Cosmos Reason v2 can help teams:

• Accelerate annotation speed by 60%
• Improve quality and efficiency in VLA
• Streamline compute and infrastructure usage 

Ensuring Data Quality and Fast Iteration

At Encord, we believe the quality of data and the speed of iteration are the two most important win criteria for AI product teams. 

Today’s best AI teams: 

1. Integrate new foundation models as soon as they are released. Across all modalities, teams that are slow to adopt or integrate new models will not be competitive. 
2. Ensure deployment repeatability by combining multiple data agents. By leveraging agents for pre-labeling, data management, and workflow automation, AI teams can focus less time building infrastructure and more time on refining their product. 
3. Continuously assess and improve annotation quality and efficiency. Just as industries are won and lost on model performance, annotation quality and efficiency will determine the rate of AI innovation across startups.  

Today’s launch marks an important step for our product and our customers in all three categories. As we move forward, we’ll continue to build out our library of Data Agents, working towards a unified data layer for all AI annotation, labeling, curation, and data management capabilities. 

In the meantime, if you’re a prospective or existing customer and would like to know more about Encord’s Data Agents, just reach out here – we’re excited to share what we’ve been working on.