Amazon Web Services (AWS) kicked off its annual re:Invent 2025 conference this week, immediately unveiling a torrent of product news centered on artificial intelligence for the enterprise. The overarching theme emphasizes empowering customers with greater control and customization over their AI agents, including a groundbreaking agent AWS claims can learn and operate independently for days.

AWS re:Invent 2025, which runs through December 5, began with a keynote from AWS CEO Matt Garman. Garman asserted that AI agents are crucial for unlocking AI's "true value," highlighting the shift from AI assistants to agents capable of performing tasks and automating processes, leading to "material business returns."

AI Agents: The Core of re:Invent

Expanded AgentCore Capabilities

AWS further enhanced its AgentCore AI agent building platform with new features. Notably, "Policy in AgentCore" allows developers to easily set boundaries for AI agents. Additionally, agents will now log and remember user interactions, and AWS introduced 13 pre-built evaluation systems to help customers assess agent performance.

Non-Stop AI Agent Workers

Three new "Frontier agents" were previewed, designed to operate autonomously. The "Kiro autonomous agent" is specifically engineered to write code, learn team workflows, and operate largely on its own for hours or days. Other new agents include one for handling security processes like code reviews and another for DevOps tasks, such as preventing incidents during code deployment. Preview versions of these agents are currently available.

Next-Gen AI Chips and Nvidia Integration

A significant hardware announcement was the new Trainium3 AI training chip, paired with the UltraServer AI system. This upgraded chip boasts impressive specifications, promising up to a 4x performance gain for both AI training and inference, alongside a 40% reduction in energy consumption.

Looking ahead, AWS also teased Trainium4, which is already in development and will feature compatibility with Nvidia's chips, signaling a strategic integration with a major industry player.

Expanding AI Model Capabilities

The Nova AI model family expanded with four new models, three designed for text generation and one capable of creating both text and images. Complementing this, AWS introduced Nova Forge, a new service enabling cloud customers to access pre-trained, mid-trained, or post-trained models. Customers can then further refine these models by training them on their own proprietary data, underscoring AWS's commitment to flexibility and customization.

Real-World Impact: Lyft's AI Agent Success

Ride-hailing giant Lyft shared its success story, utilizing Anthropic's Claude model via Amazon Bedrock to power an AI agent that addresses driver and rider inquiries. This implementation has dramatically reduced average resolution time by 87% and boosted driver usage of the AI agent by 70% this year, demonstrating tangible business benefits.

On-Premise AI Solutions: The AI Factory

Amazon also unveiled "AI Factories," a solution allowing large corporations and governments to deploy AWS AI systems within their own data centers. Developed in partnership with Nvidia, these factories integrate both Nvidia's and AWS's technologies. Customers can equip them with Nvidia GPUs or Amazon's newest homegrown AI chip, the Trainium3. This system directly addresses concerns around data sovereignty and the need for on-premise data control.

For those looking to delve deeper, attendees can explore various industry streams from re:Invent, covering topics from agentic AI and cloud infrastructure to security and much more from the flagship Amazon Web Services event in Las Vegas. A sponsored video partnership with AWS provides further insights.