AWS doubles down on agentic AI with Nova 2 and new agents
Updated on December 06, 2025 10 minutes read
Updated on December 06, 2025 10 minutes read
Nova 2 is the new generation of Amazon Nova foundation models launched at AWS re:Invent 2025. It adds four models, including Lite, Pro, Sonic and Omni, with a focus on long context, multimodal inputs and built in tools for web grounding and code execution. Compared with the first Nova release, this generation is explicitly tuned for agentic workloads and tightly integrated with Nova Forge, Nova Act and Amazon Bedrock AgentCore.
Nova 2 Lite and Nova 2 Pro include reasoning controls, web grounding and code execution so agents can both plan and act, not only generate text. Nova 2 Sonic brings real time speech to speech interaction, and Nova 2 Omni can reason across text, images, video and speech while generating both text and images. Combined with Nova Forge and AgentCore, this lets developers design agents that run for long periods, coordinate tools and stay within clear safety boundaries.
Nova Forge is a new service that lets enterprises build their own frontier models, called Novellas, using Amazon Nova checkpoints and proprietary data. Teams can combine pre trained, mid trained and post trained checkpoints with their own datasets, plus reinforcement learning gyms and distillation, to shape models around business specific tasks while keeping the underlying frontier capabilities. For enterprises that need control, performance and privacy, this is a way to move beyond simple fine tuning.
Frontier agents such as Kiro, AWS Security Agent and AWS DevOps Agent are packaged agents that target specific roles in the software lifecycle, from coding to security and operations. Nova Act is a managed service for browser automation agents powered by a custom Nova 2 Lite model, aimed at UI workflows like form filling and regression testing. Together with AgentCore, they form a layered stack where Nova models provide intelligence, AgentCore handles orchestration and governance, and agents deliver task and project-level outcomes.