Beyond the Hype: Driving Real Business Value with Agentic AI

It’s been a busy month in the world of AI agents. As we approach AWS re:Invent 2025, I’ve been reflecting on customer conversations over the past year. The shift has been remarkable—from “What is agentic AI?” to “How can we create value with it?”

Early in the gen AI wave, many organizations were stuck between impressive demos and measurable outcomes. Gen AI transformed how we search, write, and create. Agentic AI goes further, it turns intelligence into action by adding planning, reasoning, autonomy, and the ability to execute.

We believe there will soon be billions of agents across consumer, enterprise, and industrial settings, from planning trips to optimizing supply chains. Agents aren’t better chatbots; they’re digital teammates that plan, reason, and execute multi-step tasks that directly impact the bottom line. As intelligence becomes action, agentic AI finally delivers on the promise of LLMs.

AWS is committed to helping every organization and builder reimagine how work gets done with agentic AI: turning ideas into agents, code into capability, and effort into measurable impact. And we will do so by being the best place to build and deploy the world’s most useful agents. Our focus is on the foundational areas where agentic AI drives the greatest business advantage.

Accelerating Software Innovation Without Compromising Quality

Development teams are under constant pressure to deliver faster without losing quality. In just a few years, we’ve moved from traditional programming to code completion, to chat-based coding, and then to ‘vibe coding’ which is real-time collaboration between developers and AI. The next phase is AI coding agents, autonomous systems that plan, build, test, and deploy software with human oversight.

We built Kiro to make this real. Kiro transforms how developers move from concept to production. Through its multiple surfaces, from an agentic IDE to a CLI, Kiro meets developers where they are and helps them get their job done faster. Unlike assistants that stop at code generation, Kiro orchestrates multiple AI agents for planning, design, implementation, and testing.

Kiro takes your natural language prompt and turns it into clear requirements and acceptance criteria. It then analyzes your codebase and comes up with the architecture, system design, and tech stack, followed by an implementation plan with discreet tasks, sequenced based on dependencies. And Hooks automates testing and documentation, taking prototypes all the way to production. Kiro has scaled rapidly since preview, with over 250,000 developers processing trillions of tokens.

Earlier today, we announced general availability of Kiro and released new capabilities that make Kiro even more powerful: property-based testing which measures whether your code matches your specifications, a new way to checkpoint your progress on Kiro, a new CLI that brings agents directly to your terminal, and team plans with centralized management—one dashboard to control subscriptions, costs, and access across your org. We’re also making it easier for you to get started with Kiro, we are giving away one year’s worth of Kiro Pro+ for qualifying startups through the end of the year and existing AWS Activate credits now work with Kiro.

Transforming the Way We Work

Every organization faces the same issue: employees spend too much time searching for information and not enough time using it—an average of 2.5 hours a day. That’s why we built Amazon Quick Suite, not as another tool but as an intelligent teammate that understands your business context. Quick connects to your internal systems and AWS services so when you ask a question, you get answers grounded in your own data and processes.

Instead of switching between apps to gather insights, Quick unifies research, analytics, and automation in one workspace. It connects to your documents, emails, messaging services, enterprise apps like CRMs, data warehouses and the web —so you can find, analyze, and act in minutes.

Customers are turning month-long projects into days, achieving 80% time savings and 90% cost reductions. Customers like 3M, Accenture, and Principal Finance Group are already seeing driving efficiencies and seeing productivity gains.

Accelerating IT Modernization

Another resource-intensive stage of the software development lifecycle is maintaining and modernizing software. AWS Transform automates migrations with agentic AI. Instead of manual scripts, it provides self-learning workflows that adapt as they encounter new codebases—handling conversions, network translations, and framework updates.

Built on 19 years of modernization experience, Transform supports migrations for Windows, .NET, VMware, and mainframes. Customers have already saved 700,000 hours of manual effort—the equivalent to 335 developer-years. Thomson Reuters used it to convert 1.5 million lines of code per month and finish projects 4X faster than other tools.

Building Custom Solutions for Unique Needs

Every organization has workflows that are uniquely theirs. Off-the-shelf solutions can handle common use cases, but competitive advantages often come from reimagining and automating your differentiated processes with custom agents.

Building useful agents is still complex. Builders must pick the right models, frameworks, and tools to meet their accuracy and scale requirements. To simplify this, Amazon Bedrock now includes leading agentic models, such as OpenAI’s open-weight models, Anthropic’s Claude 4.5, Qwen’s Mixture-of-Experts (MoE), and Amazon Nova—all optimized for reasoning and tool use.

To make agent development easier, we launched Strands Agents, an open-source framework that uses a model-driven approach to build and run agents with just a few lines of code. Instead of rigid logic trees, Strands lets LLMs drive their own behavior, deciding when to use tools and how to adapt in real time. Since May, Strands SDK has garnered 2 million downloads and 3,000 GitHub Stars.

The next step is taking agents from POC to production. Enterprise adoption requires a secure, reliable foundation purpose-built for agents. That’s why we launched Amazon Bedrock AgentCore platform, now generally available. AgentCore provides seven fully managed modular services—Runtime, Memory, Identity, Gateway, Observability, Code Interpreter, and Browser Tool—covering the full agent development lifecycle. AgentCore works with any model, scales from zero to thousands of sessions, and delivers 99.99% availability. Randall Hunt, chief technology officer at AWS Partner Caylent said it the best in a recent interview with CRN, “Bedrock AgentCore is like taking a step back and thinking, ‘Hey, knowing what we know now, after years of working with generative AI, how would we make the perfect service for this?’ And they nailed it. They really did.”

AgentCore helps teams move from concept to production in days, not months. Druva used it to build agents that resolve 63% of support issues autonomously with 58% faster resolution. Epsilon built agents that automate campaign creation and optimization, improving ROI through personalized customer journeys.

Together, Strands + AgentCore provide a seamless path from idea to production.

Reliable Agentic Infrastructure Matters More Than Ever

As we set out to make agentic AI real for our customers, one thing became clear: infrastructure is no longer a supporting layer, but the foundation of what’s possible. From custom silicon to vertically integrated data-center design, AWS is building for the scale, latency, and reliability agentic AI demands.

I’d like to highlight two recent milestones here: Project Rainier deployed nearly half a million Trainium2 chips in under a year, powering Anthropic’s Claude models and proving how purpose-built infrastructure accelerates progress. And we announced OpenAI’s $38 billion multi-year partnership with AWS that will use EC2 UltraServers with hundreds of thousands of NVIDIA GB200 and GB300 GPUs.

As agentic AI moves from experimentation to production-scale autonomy, organizations built on deep, agent-ready infrastructure will move faster and define what comes next.

This is Just the Beginning, Join Us at re:Invent 2025 for More Agentic AI Innovation

The organizations succeeding with agentic AI share a common approach: they start with clear business problems, build on secure foundations, and scale systematically. Across Amazon and our customers, we’re seeing what’s possible when innovation meets scale and trust.

This space is incredibly exciting, and we may not know every shape this future will take, but one thing is certain: it will redefine who gets to build, create, and innovate. The next big breakthroughs will come from anyone with curiosity and an agent by their side.

I invite you to join us at re:Invent 2025 to continue this conversation. Real customer stories and lessons learned will be front and center to my keynote. We’ll introduce new agentic innovations, dive deep into the practical aspects of building and scaling production-grade agents and share insights from customers leading the transition to AI-native business. Additionally, my colleagues Asa Kalavade , Erin Kraemer , Deepak Singh , and Pasquale DeMaio will each bring unique perspectives in their innovation talks, from building reliable agents, to reimagining software development, to transforming the workplace through AI assistants. We have action packed agenda, you will learn from experts and peers and also get hands on experience on agentic AI innovations.

I hope to see you in Las Vegas this December. The next chapter of intelligent systems is only just beginning, and I can’t wait to see what we’ll build together.