Earlier this month, our CTO Alberto Gonzalez and I descended upon Las Vegas for AWS re:Invent 2025. As an AWS Partner working daily with AWS services to build real-time voice, video, and AI systems, re:Invent is more than a conference for us. It’s a chance to validate where the platform is headed and ensure we are building our clients’ applications on the most resilient, future-proof foundations possible.
Re:Invent is always a marathon of learning, networking, and walking, but this year felt different. The energy wasn’t just about “moving to the cloud” anymore; it was about the agentic era of the cloud.
Between the keynote announcements and the deep-dive sessions, it became clear that the tools we use to build real-time communication applications are evolving rapidly. We are moving from simple code completion to full-blown autonomous development, and from basic logging to AI-driven semantic observability.
Here are our key takeaways from re:Invent 2025 and what they mean for the future of voice, video, and AI applications. You can also watch this YouTube Short: Three Quick Thoughts from AWS re:Invent 2025.
The Rise of Agentic Development: Kiro and Amazon Q
If 2024 was the year of the Chatbot, 2025 is undeniably the year of the Agent. The most buzzed-about announcement was Kiro IDE, AWS’s new autonomous frontier agent for software development. The flagship feature of Kiro is its spec-driven development, capable of navigating entire codebases to plan and execute complex implementation tasks.
However, the power of agentic development isn’t limited to a single IDE. AWS’s Amazon Q Developer, brings some of these autonomous capabilities directly into other IDEs development teams already use daily, such as VS Code and JetBrains.
This means developers can leverage agentic workflows—like generating unit tests, reviewing vulnerabilities in code, or drafting documentation—without leaving their preferred environment. It effectively turns standard editors into command centers for AI-driven engineering.
Both Kiro IDE and Amazon Q Developer plugin support the Model Context Protocol (MCP). This industry standard allows teams to securely connect these AI agents to their internal data and tools.
This includes popular MCP Servers AWS provides such as:
- AWS MCP Server: For agentic access to AWS APIs and up-to-date documentation.
- AWS Knowledge MCP Server: Enabling agents to retrieve up-to-date documentation, code samples and knowledge about official AWS content
- AWS Pricing MCP Server: For AWS service pricing and cost estimates.
Responsible AI: Moving Beyond Principles to Practice
As we build more AI Voice Agents and sophisticated video analysis tools for our clients, “trust” is our most critical currency. AWS addressed this head-on with the launch of the Responsible AI Lens for the AWS Well-Architected Framework.
This practical toolkit goes beyond providing a simple checklist, it provides structured guidance on how to architect AI systems that are fair, transparent, and safe. This gives our Telehealth and EdTech clients, which often have higher data privacy and bias prevention requirements, a rigorous standard to benchmark our AI implementations against.
We are already planning to integrate these new best practices into our WebRTC assessment and development services, ensuring that when we build a Generative AI voice bot, or any AI-powered processing pipeline, it’s not just smart but also implement best practices regarding fairness, transparency and security.
Next-Gen Observability: AI and OpenSearch
One of the geekier topics we tracked was the evolution of observability tools such as Amazon OpenSearch. As real-time communication integrators, this is important because debugging a dropped call or a jittery video stream often involves hunting through a considerable amount of logs and metrics.
AWS announced that Automatic Semantic Enrichment is now available for managed OpenSearch clusters. This means we can move beyond “lexical search” (matching exact keywords like error_503) to “semantic search.” We can now query logs with natural language questions like, “Show me sessions where user quality dropped after a network handover,” and the system understands the meaning of the logs, not just the text strings.
Even more exciting is the integration of MCP Servers with OpenSearch. This allows AI agents (like Kiro or Amazon Q) to “talk” directly to our observability data. Imagine an AI agent that doesn’t just write code but can also proactively query the logs to see how that code is performing in production, effectively closing the loop between development and operations.
Other Notable Announcements at AWS Re:Invent 2025
While agents stole the show, several other announcements caught our eye for their potential impact on real-time communications:
- Amazon Nova 2: A new family of foundational models, including a new speech-to-speech model capable of ultra-low latency, paralinguistic understanding (tone/emotion), and handling interruptions naturally. This is a direct competitor to OpenAI’s Advanced Voice Mode and a powerful option for our Voice AI projects.
- Graviton5: The latest generation of AWS custom silicon offers up to 50% better performance for compute-intensive workloads. This allows for an efficient real-time processing of video and sustainable running AI inference.
- Amazon S3 Vectors: Now generally available, this simplifies building RAG (Retrieval-Augmented Generation) pipelines by allowing us to store and search vector embeddings directly next to the source media files in S3.
The Future of RTC is Agentic and Observable
The innovations we saw at re:Invent 2025 confirm that the future of real-time communications is agentic and observable.
- Faster Time-to-Market: Tools like Kiro and Amazon Q will help us prototype and deploy complex communication platforms faster.
- Safer AI: The Responsible AI Lens ensures we are building solutions that enterprise compliance teams can trust.
- Smarter Operations: AI-driven OpenSearch means we can detect and fix quality issues before users even report them.
We left Las Vegas exhausted but inspired. The convergence of RTC and Agentic AI is accelerating, and we can’t wait to apply these new tools to your projects in 2026.
Further reading on the WebRTC.ventures blog:
- Voice + Action: The Convergence of WebRTC, Conversational AI, and Agentic Systems
- How to Choose Voice AI Agent Patterns: Conversation-based vs Turn-based Design
- 3 Ways to Deploy Voice AI Agents: Managed Services, Managed Compute, and Self-Hosted
- Our Clients Succeed: AVA Intellect, Built with WebRTC.ventures, Acquired by Wowza
- How to Build Voice AI Applications: A Complete Developer Guide
- Deploying AI Agents in Production with AWS (AgilityFeat blog)
- Five Tips for Integrating LLMs into WebRTC Applications (white paper)
