I attended the AWS Summit Amsterdam 2026 keynote on May 27, and the announcements were massive. From a new agentic AI platform to OpenAI models running on AWS infrastructure, here is everything that matters for enterprise builders.
The Keynote: Agentic AI Takes Center Stage
The keynote, hosted by Martin from AWS, focused on a single theme: AI agents are transforming what is possible, and AWS wants to make it easier than ever to build, deploy, and scale them.
The session (KEY101) was structured around three pillars:
- Infrastructure — custom silicon and S3 as the universal data layer
- Data — open architecture with catalog and semantic layers
- Inference — taking AI from prototype to production
Amazon S3: The Universal Data Foundation
AWS positioned S3 as the starting point for everything — not just object storage, but a unified platform supporting:
- Object storage (classic S3)
- Tabular data (Iceberg, Spark)
- Metadata management
- Vector storage for AI/ML
The integrated ecosystem includes Analytics, Database, AI/ML, Search, and Streams — with Iceberg, Apache Spark, PostgreSQL, OpenSearch, and Kafka all running on top of S3.
Amazon S3 Files (NEW)
A new service providing fully-featured, high-performance file system access to S3 data. You can mount any S3 bucket inside EC2, Lambda, ECS, EKS, or Fargate — eliminating the need for separate file storage solutions.
Open Data Architecture
AWS introduced a Catalog and Semantic Layer sitting above the service tiers, providing metadata, business context, governance rules, and semantic meaning across all data assets.
Amazon Bedrock: Broadest Model Choice
Bedrock now offers the broadest choice of models from 18+ providers:
- AI21Labs (Jamba), Amazon (Nova), Anthropic (Claude Opus, Sonnet, Haiku)
- Cohere (Command, Embed, Rerank), DeepSeek, Google (Gemma)
- Moonshot AI (Kimi), Luma (Ray), Meta (Llama), MiniMax
- Mistral (Mistral, Mixtral, Pixtral), NVIDIA (Nemotron)
- OpenAI (GPT, GPT-O3S, GPT-O3S-Safeguard), Qwen (Qwen3-Next, Qwen3-VL)
- Stability AI (Stable Diffusion, Stable Image), TwelveLabs (Marengo, Pegasus), Writer (Palmyra)
The platform provides built-in security, customization, guardrails and policy checks, agentic AI capabilities, and models and infrastructure — all designed to take you from prototype to production reliably, securely, and at scale.
“The power to choose. We don’t force you into a single fixed path.”
Amazon Bedrock AgentCore (NEW)
The biggest announcement: Amazon Bedrock AgentCore — an agentic platform to build, deploy, and operate agents securely at scale.
The Problem
“Getting agents to production is still difficult.” The journey from Logic → Context → Secure → Deploy → Audit → Scale requires solving many undifferentiated challenges. AgentCore lets you focus on innovation rather than operations.
Nine Components
| Component | Purpose |
|---|---|
| Runtime | Agent execution environment |
| Memory | Persistent agent state |
| Gateway | API management and routing |
| Identity | Authentication and authorization |
| Observability | Native monitoring and tracing |
| Code Interpreter | Safe code execution |
| Managed Browser | Web interaction capabilities |
| Evaluations | Quality and safety testing |
| Agent Registry (NEW) | Catalog and discover agents |
Amazon Bedrock Managed Agents — Powered by OpenAI (NEW, Preview)
“Frontier intelligence meets enterprise-grade infrastructure.”
This is a landmark partnership: OpenAI models (including GPT 5.5) running inside Amazon Bedrock with full governance and observability.
Runtime layer:
- Agentic Loop, Inference, Memory, Skills (OpenAI Harness)
Environment layer:
- Compute, Storage, Networking, Security, Tools/MCP
Foundation:
- OpenAI Models (including GPT 5.5)
- Amazon Bedrock Governance and Observability
This means enterprises can now use the most capable OpenAI models while keeping AWS security, compliance, and infrastructure guarantees.
Customer Story: EDSN — Dutch Energy Data
EDSN (energiedata voor iedereen — energy data for everyone) shared their AWS journey:
- 91% cost reduction compared to on-premises infrastructure
- 500x built-in data growth capacity for the next decade
- Building AI capabilities on this S3 foundation to power the Dutch energy transition
Customer Story: Signify (Philips Hue)
Signify, the company behind Philips Hue, chose Bedrock AgentCore to power a new AI assistant in the Hue app. Users can create personalised lighting scenes simply by typing or speaking a mood, location, or style.
Results since starting in 2024:
- 20% improved agent response time
- 27% optimised costs for overall AI use case
Customer Story: KBC Bank — Kate, the AI Banking Agent
Karen Van De Woestyne, General Manager Transform and Ecosphere at KBC Bank and Verzekering, presented perhaps the most impressive enterprise AI deployment at the summit.
Three Shifts That Define KBC’s Future
Karen outlined three fundamental shifts driving KBC’s transformation:
- Transactions → Journeys — moving from isolated interactions to supporting complete life paths
- Asking → Anticipating — from reactive responses to proactive suggestions
- Bank → Digital Buddy — from financial institution to trusted companion
“This is not just a story about my daughter Amber, 27, dreaming of leaving the home.”
KBC serves 6.1 million customers and is shifting from reactive banking to proactively supporting life journeys — not just transactions.
Intent Thinking: The Engine Behind Kate
“Intent Thinking turns signals into trusted action, so Kate can anticipate the next meaningful step.”
KBC built what they call an Intent Engine — a system that interprets customer signals (browsing patterns, life events, transaction history) and converts them into proactive, personalised recommendations. This is what powers Kate’s ability to suggest the right financial product at the right moment.
The Roadmap: From Now to the New Bliss
- 2025: BUILDING — establishing the AI foundation
- 2026: ACCELERATE — scaling across customer journeys
- 2027: AMBITION — full anticipatory banking
Kate: 80 Million Conversations
Kate is KBC’s trusted digital buddy, built on AWS:
- 80M conversations started by customers
- 25K+ customer queries answered every day
- 350+ FTE equivalent operational output
- 50+ third-party services integrated
- 400K+ products and services sold per year
Built on: Amazon Bedrock AgentCore, Amazon EKS, Amazon S3
The Ecosphere: Beyond Banking
Kate operates inside what KBC calls The Ecosphere — a customer universe extending beyond traditional banking:
- Myhome: 40K customers onboarded, 138K questions answered
- Mymobility: 340K customers onboarded, 1.5M+ train tickets issued, 97 cars sold
“If we miss Amber’s intent, someone else will own the moment, shape the relationship, and capture the value.”
The Full Platform: “The Best Place to Build”
The keynote closed with the full AWS AI platform vision — six pillars working together:
- Kiro — AWS’s new AI-native IDE, reimagining how we write code
- Agents — Amazon Bedrock AgentCore for production deployment
- Open Data Architecture — catalog, semantic layer, governance
- Inference — Amazon Bedrock with 18+ model providers
- Infrastructure — AWS Graviton5 custom silicon
- Data — S3 as the universal foundation
- Amazon Quick and AWS Transform — business intelligence and migration tools
“It looks like when you have the intelligence and scale to back it up.”
Key Takeaways
- S3 is no longer just storage — it is the universal data platform spanning objects, tables, metadata, and vectors
- AgentCore solves the production gap — the hardest part of agents is not building them, it is operating them at scale
- OpenAI on Bedrock is real — GPT 5.5 with enterprise guardrails changes the competitive landscape
- Agents deliver measurable ROI — KBC’s 350 FTE equivalent and Signify’s 27% cost reduction prove the business case
- The power to choose — AWS is betting on model diversity, not lock-in
My Take
The theme of this keynote was clear: agentic AI is the next platform shift, and the companies winning are those who have already moved past experimentation into production deployment.
KBC’s numbers are staggering — 80 million conversations, 350 FTE equivalent output. That is not a pilot. That is a transformation.
For infrastructure architects evaluating their AI platform strategy, the combination of AgentCore (operations) + Managed Agents (frontier models) + S3 (unified data) represents a compelling full-stack offering.
I attended AWS Summit Amsterdam 2026 on May 27. For more cloud infrastructure and AI content, explore my Kubernetes workshops or book a consultation.
