Introduction – A New Era of Artificial Intelligence
AWS re:Invent 2025 reinforced a movement that had already been gaining momentum in the market: the definitive transition of generative Artificial Intelligence toward agentic solutions capable of executing complex tasks, making decisions within defined rules, and operating in an integrated way with corporate systems.
With more than 60,000 in-person attendees, millions of online viewers, and hundreds of announcements, the event made it clear that AWS’s focus goes far beyond language models. The spotlight now is on how to turn AI into real business value, with governance, security, automation, and integration with the corporate ecosystem.
In this article, I highlight four strategic launches presented at re:Invent 2025 that have direct potential to impact companies of all sizes—especially those seeking productivity, operational efficiency, security, and continuous innovation.
Amazon Bedrock AgentCore and Amazon Nova
Amazon Bedrock is AWS’s managed platform for building generative AI applications, enabling the use of multiple foundational models through a single API, with a focus on security, scalability, and corporate governance. With AgentCore, Bedrock has evolved into a true operating system for AI agents.
AgentCore is a comprehensive suite of services for building, running, managing, and monitoring AI agents at enterprise scale. Although AgentCore itself was not a new launch, it received important capabilities that strengthen its use in corporate environments. Highlights include:
- AgentCore Policy, which allows control and auditing of agent actions through natural language policies
- AgentCore Evaluations, which makes it possible to continuously evaluate the quality of responses and decisions made by agents in production, bringing governance and clear metrics to AI-based applications
- AgentCore Memory now includes Episodic Memory, allowing agents to learn and adapt based on experience, accumulating knowledge over time
- AgentCore Runtime now offers bidirectional streaming for natural conversations in which agents listen and respond while handling interruptions
- AgentCore Identity now supports custom claims for enhanced authentication rules in multi-tenant environments
Key capabilities of AgentCore include:
- Serverless, framework-agnostic agent execution
- Persistent memory management
- Agent identity and authentication control
- Secure code execution and controlled web browsing
- Central gateway for tool calling
- Natural language policies defining what an agent can or cannot do
- Full observability with metrics, logs, and traceability
- Continuous evaluation of agent quality and behavior
All this is compatible with frameworks such as LangGraph, CrewAI, LlamaIndex, Strands Agents, and supports the MCP (Model Context Protocol) and Agent-to-Agent (A2A) standards.
Amazon Nova Foundation Models
AWS announced the evolution of the Amazon Nova Foundation Models family, reinforcing its strategy to offer proprietary models with high performance, multimodality, and excellent cost-efficiency. The new models significantly expand reasoning, context, and voice interaction capabilities and add native features such as web grounding, tool use, and code interpretation, making it easier to create more complete and trustworthy AI applications.
Amazon Nova Foundation Models:
- Nova 2 Lite – fast and cost-effective reasoning for everyday workloads
- Nova 2 Pro – the most advanced model for complex tasks and sophisticated agents
- Nova 2 Omni – a fully multimodal model (text, image, audio, and video)
- Nova 2 Sonic – real-time voice interaction with Portuguese support
Amazon Nova Forge
AWS released Amazon Nova Forge, a new service that enables organizations to create their own customized models by combining proprietary data with Nova checkpoints. This means companies can train models tailored to their business domain, preserving fundamental reasoning capabilities while incorporating organization-specific knowledge—a major advantage for industries requiring accuracy, specialization, and competitive differentiation.
This combination of Nova 2 models with the ability to build customized variants expands the potential for AI adoption, reduces development complexity, and accelerates business problem-solving through AI more aligned with corporate needs.
These launches solve three major challenges companies face:
- Scaling AI with governance
- Reducing operational and security risks
- Transforming POCs into production-ready solutions
Reference links
Amazon Bedrock AgentCore
https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-agentcore-is-now-generally-available/
Amazon Nova Models
https://aws.amazon.com/nova/models/
Amazon Nova Forge
https://aws.amazon.com/blogs/aws/introducing-amazon-nova-forge-build-your-own-frontier-models-using-nova/
Kiro
AWS Kiro, one of AWS’s major announcements in 2025, is a development platform (IDE) based on Visual Studio Code designed to redefine how applications are designed, developed, and maintained with the support of Artificial Intelligence. Unlike other tools in the market focused only on code generation, Kiro was built to work with complete software context—connecting requirements, architecture, business rules, and implementation within the same environment. Kiro’s major differentiator lies in its spec-driven approach, as well as mechanisms like Hooks, Steering, and more. At re:Invent, new capabilities were introduced: Powers and Autonomous Agent.
Kiro Powers
Powers are reusable bundles of knowledge and automation that include:
- Technical and business context
- Best practices
- Integrations via MCP
- Automated actions
- Expert and ISV guidance
These bundles can be created by internal teams, partners, or ISVs and shared across projects, ensuring standardization, governance, and development acceleration.
In practice, Kiro Powers allows companies to transform scattered knowledge into reusable assets, reducing dependence on specific individuals and increasing delivery consistency.
Kiro Autonomous Agent
The Kiro Autonomous Agent represents the evolution from assisted to autonomous development. It acts as an AI software engineer capable of executing tasks asynchronously and continuously, without relying on interactive sessions.
Capabilities include:
- Durable memory and continuous learning
- Execution of complex tasks end-to-end
- Operation across multiple repositories simultaneously
- Automatic planning, execution, and validation
- Operation in customized environments
This model enables teams to delegate activities such as fixes, refactoring, improvements, and technical analysis to autonomous agents, freeing professionals for higher-value strategic work.
Reference links
Kiro Powers
https://kiro.dev/blog/introducing-powers/
Kiro Autonomous Agent
https://kiro.dev/blog/introducing-kiro-autonomous-agent/
AWS DevOps Agent (Preview)
The AWS DevOps Agent is an AI agent focused on operational excellence and system reliability. It acts as a virtual SRE engineer capable of analyzing environments in real time and responding quickly to incidents.
The service integrates natively with observability tools and CI/CD pipelines, enabling:
- Automatic incident detection
- Correlation across metrics, logs, and traces
- Analysis of recent code or deployment changes
- Root cause identification
- Recommendations and corrective actions
- Learning from historical incidents
The goal of the DevOps Agent is to drastically reduce MTTR (Mean Time To Resolution) and increase operational stability, especially in complex and distributed environments.
The AWS DevOps Agent stands out for its ability to operate in heterogeneous environments, integrating natively with major observability and automation tools. It consumes telemetry signals and acts directly within workflows through:
Observability: Amazon CloudWatch, Datadog, and Dynatrace
Automation & CI/CD: GitHub Actions and GitLab CI/CD
Incidents remain one of the biggest invisible costs in IT:
- Downtime
- Customer impact
- Revenue loss
- Team overload
Reference links
AWS DevOps Agent
https://aws.amazon.com/devops-agent/
Launch blog
https://aws.amazon.com/blogs/aws/aws-devops-agent-helps-you-accelerate-incident-response-and-improve-system-reliability-preview/
AWS Security Agent
The AWS Security Agent is an Artificial Intelligence agent designed to operate continuously throughout the entire security lifecycle—from design to operation. It enables security to move from being just a final step to an integrated process from the beginning.
The agent works across multiple fronts:
- Definition of security requirements
- Architecture validation
- Code review with a security focus (SAST)
- Dynamic application testing (DAST)
- On-demand pentesting
- Vulnerability remediation guidance
All integrated into the development workflow. Security is no longer just a final stage—it has become a continuous process. The Security Agent enables the Shift Left Security model, reducing risks and costs.
Benefits
- Security integrated into DevOps
- Reduced vulnerabilities in production
- Automation of audits and compliance
- Improved corporate security posture
Reference links
AWS Security Agent
https://aws.amazon.com/security-agent/
Blog
https://aws.amazon.com/blogs/aws/new-aws-security-agent-secures-applications-proactively-from-design-to-deployment-preview/
Conclusion
The major turning point this year is autonomy. We moved from passive language models to active agents capable of making decisions based on business rules. These four launches we discussed are the pillars of this new phase: they reduce friction between idea and execution. For companies, the real impact will be measured by operational efficiency and the capacity to innovate continuously without the constraints of manual processes.
In summary, re:Invent 2025 established AI as the central engine of modern efficiency. We are no longer talking about promises but about tools ready to run the core business of large organizations. The question for 2026 is not whether your company should adopt AI, but which complex tasks you will delegate to your agents today to ensure that your operation leads tomorrow. The infrastructure is ready; execution now depends on strategy.
Numen supports its clients throughout the entire AI adoption journey—from strategy and architecture to implementation and governance.
If your company wants to understand how to apply these launches in practice, reduce risks, and generate real value with AI, get in touch: https://numenit.com/contact-us/
Fernando Pena is Cloud Director at Numen in the United States, Cloud Solutions Architect, AWS Certified Specialist, and was a speaker at AWS re:Invent 2025 in Las Vegas, Nevada.
