AWS continues to unveil groundbreaking innovations at re:Invent 2025, with announcements spanning frontier agents that can work autonomously for days, an expansion of the Amazon Nova model family, the availability of Trainium3 UltraServers—delivering up to 4.4x more compute performance than the previous generation—and the launch of AWS AI Factories for dedicated AI infrastructure in customers' existing data centers.
Simplifying purpose-built AI infrastructure with Amazon Bedrock AgentCore
Amazon Bedrock AgentCore is the most advanced platform for building and deploying agents securely at scale. To move agents from prototype to production, companies need infrastructure that is secure, reliable, scalable, and purpose-built for the non-deterministic nature of agents. Agents need a foundation that scales dynamically, supports long-running workloads, and allows them to store and retrieve context instantly and securely. Today, early adopters are diverting significant resources to build this infrastructure from scratch, a labor-intensive and time-consuming process that slows down development cycles.
Amazon Bedrock AgentCore addresses these challenges by offering essential, fully managed services. AgentCore supports any framework (like CrewAI, LangGraph, LlamaIndex, Google ADK, OpenAI Agents SDK, and Strands Agents) or model while handling critical agentic AI infrastructure needs. In just five months since preview, organizations including Amazon Devices Operations & Supply Chain, Cohere Health, Cox Automotive, Heroku, Natera, MongoDB, PGA TOUR, Pulumi, Thomson Reuters, Workday, Snorkel, and Swisscom are already using AgentCore to build agents, and developers have downloaded it more than 2 million times.
Bedrock AgentCore customer momentum
PGA TOUR, a pioneer and innovation leader in sports, has built a multi-agent content generation system to create articles for its digital platforms. The new solution, built on AgentCore, enables the PGA TOUR to provide comprehensive coverage for every player in the field by increasing content writing speed by 1,000% while achieving a 95% reduction in costs.
MongoDB, a database platform, leveraged AgentCore to reshape how it designed and operationalized AI within the company. Through AgentCore's implementation, the company eliminated weeks of evaluation cycles and consolidated multiple disparate tools into a single, production-ready solution. By seamlessly integrating MongoDB's AWS infrastructure and utilizing MongoDB Atlas as the embedded Knowledge Base for Amazon Bedrock, its development teams deployed an agent-based application in just eight weeks.
This process previously took months of infrastructure work and continuous maintenance. This streamlined approach enabled MongoDB to scale its AI initiatives with greater accuracy, contextual awareness, and consistency, while significantly reducing manual overhead.
Swisscom, Switzerland's leading telecoms provider, selected AgentCore to deploy containerized agents with AgentCore Runtime for scalable hosting, AgentCore Identity for seamless authentication across systems, and AgentCore Memory for tracking customer interactions. By standardizing how agents are built, deployed, and integrated, Swisscom now has an enterprise-grade foundation that lets teams focus on business problems instead of infrastructure. With AgentCore and Strands, the company launched their business-to-consumer agent solution in just four weeks, focusing on personalized sales assistance and automated technical support.
AWS simplifies model customization to help customers build faster, more efficient AI agents
Running AI applications at scale remains expensive and resource-intensive, particularly for AI agents that spend significant time on routine tasks that don't require advanced intelligence. AWS is announcing new Amazon Bedrock and Amazon SageMaker AI capabilities that make advanced model customization accessible to any developer. Reinforcement Fine Tuning (RFT) in Amazon Bedrock simplifies the model customization process, delivering 66% accuracy gains on average over base models, with customers like Salesforce demonstrating up to 73% improvement in accuracy over base models. Amazon SageMaker AI now supports serverless model customization capabilities that accelerate workflows from months to days, with customers like Collinear AI cutting experimentation cycles from weeks to days.
Kiro powers: Access specialized expertise to accelerate software development
As developers increase their use of AI agents for a wider range of software development tasks, they want agents that have deep knowledge of the tools they use every day and that are specialized in their workflows, like user interface or application programming interface development. Kiro powers enable developers to give Kiro agents instant expertise in these tools and workflows in a single click. Powers can be comprised of a combination of MCP servers for specialized tool access, steering files with best practices, and hooks to trigger specific actions helping developers equip Kiro agents with workflow-specific knowledge spanning the application lifecycle: design, development, deployment, and observability. By loading only when needed, powers help developers work with efficient token usage, precision, and speed.
Developers can build with expertise in their everyday tools using Kiro powers from Datadog, Dynatrace, Figma, Neon, Netlify, Postman, Stripe, Supabase, and AWS—with more to come. Developers can also create and share their own powers with the community.
Checkpointless training on SageMaker HyperPod: recover from model training faults in minutes
Amazon SageMaker HyperPod simplifies infrastructure management for model training and deployment, reducing costs by up to 40%. As training scales across hundreds or thousands of accelerators, faults like hardware or software failures can occur. Traditional checkpoint-based recovery can take up to an hour, which is expensive, consumes storage, and leaves multi-million-dollar compute clusters idle during recovery. AWS is announcing checkpointless training on SageMaker HyperPod—automatically recovering from infrastructure faults in minutes with zero manual intervention, enabling training cluster efficiency of up to 95% on clusters with thousands of AI accelerators.
Checkpointless training continuously preserves model state across the training compute cluster. When faults occur, the system automatically swaps out faulty components and recovers training using a peer-to-peer transfer of model and optimizer states from nearby healthy accelerators—mitigating lengthy downtime so teams can focus on building the best AI model for their use case.
Strands Agents SDK now in Typescript (preview)
AWS is bringing Strands Agents, the open source, model-driven, AI agent framework to TypeScript, one of the world’s most popular programming languages and communities. Developers love TypeScript because it catches errors early and provides powerful tooling while still letting them write familiar, flexible JavaScript. Strands provides full support for key TypeScript features, including type safety, async/await, and modern JavaScript/TypeScript patterns. Originally available in Python with over 3M downloads, AWS is extending Strands Agents to give developers the ability to build their entire agentic stack in TypeScript using the AWS CDK. Head to the Strands Agents Github to join the millions of developers who are building today.
Strands adds support for edge devices
AWS is announcing the general availability of Edge Device support for Strands Agents. With edge device support, customers can use the Strands Agents SDK to create autonomous AI agents that can run on small-scale devices, unlocking new agentic use cases in automotive, gaming, and robotics. Developers can also implement bi-directional streaming capabilities and run agents using local models like Ollama and Llama.cpp. Head to the Strands Agents Github to join the millions of developers who are building today.
Amazon expands Nova family of models and pioneers “open training” with Nova Forge
Amazon is expanding its Nova portfolio with four new models that deliver industry-leading price-performance across reasoning, multimodal processing, conversational AI, code generation, and agentic tasks. Nova Forge pioneers "open training," giving organizations access to pre-trained model checkpoints and the ability to blend proprietary data with Amazon Nova-curated datasets.
Nova Act achieves breakthrough 90% reliability for browser-based UI automation workflows built by early customers. Companies like Reddit are using Nova Forge to replace multiple specialized models with a single solution, while Hertz accelerated development velocity by 5x with Nova Act.
AWS unveils 3 frontier agents, a new class of AI agents that work as an extension of your software development team
Frontier agents represent a step-change in what agents can do. They’re autonomous, scalable, and can work for hours or days without intervention. AWS is announcing three frontier agents—Kiro autonomous agent, AWS Security Agent, and AWS DevOps Agent. Kiro autonomous agent acts as a virtual developer for your team, AWS Security Agent is your own security consultant, and AWS DevOps Agent is your on-call operational team. Companies including Commonwealth Bank of Australia, SmugMug, and Wester Governors University that have used one or more of these agents to transform the software development lifecycle.
Trainium3 UltraServers enable customers to train and deploy AI models faster at lower cost
As AI models grow in size and complexity, training cutting-edge models requires infrastructure investments that only a handful of organizations can afford. Amazon EC2 Trn3 UltraServers, powered by AWS's first 3nm AI chip, pack up to 144 Trainium3 chips into a single integrated system, delivering up to 4.4x more compute performance and 4x greater energy efficiency than Trainium2 UltraServers. Customers achieve 3x higher throughput per chip while delivering 4x faster response times, reducing training times from months to weeks.
Customers including Anthropic, Karakuri, Metagenomi, NetoAI, Ricoh, and Splash Music are reducing training and inference costs by up to 50% with Trainium, while Decart is achieving 4x faster inference for real-time generative video at half the cost of GPUs, and Amazon Bedrock is already serving production workloads on Trainium3.
𝐒𝐭𝐚𝐲 𝐢𝐧𝐟𝐨𝐫𝐦𝐞𝐝 𝐰𝐢𝐭𝐡 𝐨𝐮𝐫 𝐥𝐚𝐭𝐞𝐬𝐭 𝐮𝐩𝐝𝐚𝐭𝐞𝐬 𝐛𝐲 𝐣𝐨𝐢𝐧𝐢𝐧𝐠 𝐭𝐡𝐞 WhatsApp Channel now! 👈📲
𝑭𝒐𝒍𝒍𝒐𝒘 𝑶𝒖𝒓 𝑺𝒐𝒄𝒊𝒂𝒍 𝑴𝒆𝒅𝒊𝒂 𝑷𝒂𝒈𝒆𝐬 👉 Facebook, LinkedIn, Twitter, Instagram
