Do you remember that incredible exhilarating moment when you first built something and realized you could create anything you wanted? That’s the powerful human feeling Dr. Swami Sivasubramanian, VP of Agentic AI at AWS, put into words during his keynote at AWS re:Invent 2025. He took us back to his own childhood moment of triumph, writing a scientific calculator program to highlight the massive change happening right now. 

The biggest takeaway for us at Cloudelligent is this: WHO gets to build is changing forever. We are officially moving past the gatekeeping of complex code endless API calls and dense syntax. Dr. Swami’s message is clear! AI Agents give all of us dreamers, thinkers, and builders the freedom to imagine and create without limits. 

We are thrilled to dive into some of the groundbreaking Agentic AI announcements this year that make this freedom a reality. 

Advancement in Strands Agents

Dr. Swami shared that the open-source Strands Agents SDK has long given developers the freedom to build without limits, and now that ability is moving even faster. Strands Agent simplifies the entire agent creation process. It embraces a powerful model-driven approach. This means developers no longer have to struggle writing thousands of lines of complex orchestration code. Instead, the LLM autonomously handles all the planning and execution.  

Our teams at Cloudelligent are thrilled to see this open-source tool eliminate complexity and truly empower developers. And the adoption proves it. Strands has been downloaded more than 5 million times! 

This week, AWS announced two exciting new capabilities: 

• TypeScript Support: The feature extends Strands to one of the world’s most popular programming languages. Developers now have the choice between the Python and TypeScript frameworks. Benefits include full type safety and modern JavaScript patterns, making it easy to build full-stack agent applications. 
• Edge Device Support: Strands Agents can now run on small edge devices using local models like llama.cpp. This makes it possible to run agents closer to data in fields such as automotive, gaming, and robotics. We can now build more responsive applications. 

AgentCore Memory Episodic Functionality

One of the biggest challenges with AI and POC projects is that they often aren’t designed to be production ready. Many great ideas get stuck in what Dr. Swami calls the “Proof of Concept Jail,” never fully realizing their potential. That’s why Amazon Bedrock AgentCore was introduced which provides a robust platform to build, deploy, and operate agents securely at scale. 

He implored us to think about what makes an agent truly intelligent: Memory. Think about your favorite local coffee shop. What makes them exceptional? They remember you, your favorite drink, and maybe even your work schedule! Agents need that level of context. 

While AgentCore already handles short-term conversation context and long-term general insights, the missing piece was the ‘when’ and the ‘why’ behind user behavior. And thus, AWS announced AgentCore Memory Episodic Functionality. It allows agents to store and recall specific interactions as discrete experiences, similar to how we remember a particular event in our lives. 

The more your agents experience, the smarter they become! 

Reinforcement Fine-tuning in Amazon Bedrock

Dr. Swami tackled a big compromise for builders: settling for generic models or spending millions on complex, advanced customization. The solution is removing that complexity and cost entirely! 

He introduced Reinforcement Fine-tuning (RFT) in Amazon Bedrock, making this advanced model customization accessible to every developer. The best part? No PhD required! RFT completely automates the complex pipeline. All you do is select a base model, feed it your Bedrock logs, and choose a reward function. That simple process delivers up to 66% accuracy gains on average. You get to use smaller, faster, and more affordable models while getting better quality tailored exactly to your agent’s tasks. 

Model Customization on Amazon SageMaker AI

While Amazon SageMaker AI has always given us the tools to build and deploy custom models, the process was complicated. As Dr. Swami reminded us, it involved weeks of picking criteria, cleaning messy data, tweaking hyperparameters, and then watching loss curves break at 2 AM. The journey from idea to business value took months! 

That changes now. AWS has released new serverless Model Customization in Amazon SageMaker AI which compresses that journey down to days. They offer two simple experiences: a self-guided approach for experts, or an amazing agent-driven experience where an AI expert guides you through the full workflow. Simply use natural language to explain your use case, and the agent handles everything. It can recommend fine-tuning techniques, generate synthetic data, set up serverless infrastructures, and evaluate the final model. Any heavy lifting is removed, so your team can focus on the outcome. 

Checkpointless Training on Amazon SageMaker Hyperpod

Dr. Swami spent time on the infrastructure side, too. He reminded us that building big models often leads to frustrating cluster failures. The traditional fix for a fault meant pausing the entire cluster, diagnosing the issue, and restoring from a checkpoint. That leaves expensive AI resources sitting idle for hours, which is a huge headache and a major cost drain! 

The brilliant solution is Checkpointless Training on Amazon SageMaker HyperPod. This feature is a paradigm shift for training resiliency. Instead of rolling back to an old checkpoint, HyperPod continuously preserves the model state across the distributed cluster. When a fault happens, it smoothly swaps out the faulty hardware and instantly grabs the model state from healthy accelerators. The result? Recovery happens in minutes with zero manual work, saving you money and maximizing your powerful AI compute resources. 

Amazon Nova Act

“Now, the future of Agentic AI is in agents that can do anything. It’s agents we can rely on to do everything.” – Dr. Swami Sivasubramanian 

When agents move from prototypes to automating production work, they need to be highly reliable. As Dr. Swami pointed out, older automation tools like RPA were rigid and broke easily. Basic LLMs were error-prone because they lacked tight control. Businesses just needed reliable automation that was simple! 

That need is being met with the availability of Amazon Nova Act. We’re really excited about this new service which is built to manage fleets of agents that automate production UI workflows with exceptional reliability. The VP of Agentic AI mentioned achieving 90% success in enterprise settings! 

Nova Act works by tightly integrating the model, the orchestrator, and the actuator, training them together like a body and brain learning to walk simultaneously. It even uses specialized Reinforcement Learning Gyms replicas of enterprise UIs so the agents can learn from millions of trials without interfering with your live systems. The entire approach ensures high reliability and fast time-to-value for complex automation tasks. 

Agentic Self-service Capabilities in Amazon Connect

“Over the next few years, human AI teams will fundamentally rewire how work gets done.” – Colleen Aubrey 

Shifting the focus from core development tools, Colleen Aubrey who is the SVP of Applied AI Solutions at AWS, took the stage to remind us that the real prize of AI is more than just reduced effort. The goal is actually unlocking better customer experiences. She believes Agentic teammates are about to fundamentally transform how companies deliver service. AWS is putting agents to work right inside Amazon Connect, its cloud-based customer service application. 

Connect is introducing powerful agentic self-service capabilities that allow AI agents to understand, reason, and take action across voice and messaging. Cloudelligent is excited to see how these advancements will allow businesses to automate complex support while ensuring seamless escalation to a human representative. The focus is truly about making people and AI agents partners. 

Here are the eight new features: 

• AI Agent Assistance: Real-time help for human agents during customer calls. 
• AI Agent Observability: Tools to monitor agent performance and behavior. 
• AI Case Summarization: Automated generation of concise interaction summaries. 
• Flow Modules as LLM Tools: Using existing contact center logic as tools for AI agents. 
• MCP Tool Support: New support for Management Console Portals (MCP) tools. 
• Nova Sonic Speech-to-Speech Integration: Natural, adaptive voice interactions that match customer tone. 
• Real-time Agentic Recommendations: Automatic suggestions for the agent’s next best step. 
• Self-Service Evaluation: Tools to systematically measure the quality of AI-driven self-service. 

Build Without Limits in the New Era of Agentic AI with Cloudelligent

“We are making it easier for anyone to build and use these agents.” – Dr. Swami Sivasubramanian 

Dr. Swami captured it perfectly. That rush of creation you felt the first time you programmed something is no longer a memory. Agentic AI makes that feeling real every day. The freedom to move from idea to impact is tangible and faster than ever. 

The Cloudelligent team is excited to help our clients harness this very moment. With Strands Agents, Amazon Bedrock AgentCore, Amazon Nova Act, Amazon SageMaker AI, and Amazon Connect, we guide teams to simplify complexity, scale smarter, and accelerate outcomes. The tools are ready, the possibilities are limitless, and the future of building has arrived. 

Curious how agentic AI fits into your roadmap? Book a FREE AI/ML assessment with us today.