The spring of 2026 has been all about one word: agents. From April’s “What’s Next with AWS” event through the global wave of AWS Summits — with New York as the flagship — AWS laid down its vision of an agentic future and shipped the infrastructure to back it up. Lambda MicroVMs give AI-generated code a secure sandbox to play in, Graviton5 brings 192 cores of Arm muscle to the party, and services like AWS Context and Continuum aim to make enterprise agents both smarter and safer. Meanwhile, Amazon Quick evolved from a personal AI assistant into a full-blown autonomous agent platform, and OpenAI models landed in Bedrock. Let’s dive in.
We enjoyed the style of our Summer 2025 edition so much that we’re keeping the tradition alive. Grab a coffee — these announcements deserve it.
Table of contents
- Compute & Silicon
- Serverless
- AI & Agents
- Security
- Storage
- Developer Experience
- Operational Excellence
Compute & Silicon
Amazon EC2 M9g and M9gd instances – AWS Graviton5 goes GA
The chip everyone’s been waiting for finally earned its driver’s license! Graviton5 rolls up with 192 cores, a four-chiplet architecture, DDR5-8800 memory, and PCIe gen6 — all baked at 3nm. The result? Up to 25% better compute performance, 35% faster web apps, 35% faster ML inference, and 30% faster databases compared to Graviton4. It’s like AWS put a Formula 1 engine in your general-purpose workloads and somehow made it more fuel-efficient at the same time.
Posted on: Jun 10, 2026 | Read more
Amazon EC2 C9g and C9gd instances – Graviton5 for compute-intensive workloads
Not content with just the general-purpose crowd, Graviton5 crashed the compute-optimized party too. C9g instances bring that same 192-core beast to video encoding, scientific modeling, and distributed analytics. With the fastest cloud memory available (DDR5 8800MT/s), your number-crunching workloads now have a best friend who never forgets and never slows down.
Posted on: Jun 30, 2026 | Read more
Amazon EC2 G7 instances – NVIDIA RTX PRO 4500 Blackwell GPUs
AWS became the first major cloud provider to strap NVIDIA’s RTX PRO 4500 Blackwell Server Edition GPUs to its instances. The numbers speak for themselves: 4.6x AI inference performance and 2.1x graphics performance compared to G6 instances. Your generative AI workloads just went from “please wait” to “already done” territory.
Posted on: Jun 18, 2026 | Read more
Serverless
AWS Lambda MicroVMs – A new compute primitive is born
This might be the most important serverless announcement since Lambda itself. Lambda MicroVMs is an entirely new compute primitive built on Firecracker that gives you VM-level isolation (no shared kernel!), near-instant launch and resume, state preservation for up to 8 hours, and full lifecycle control — all without managing infrastructure. The killer use case? Running untrusted code from users or AI agents in a sandbox that’s actually secure. Your agents can now generate and execute code without you losing sleep over what they might touch. It’s serverless meets hardware-grade isolation, and it changes how we think about running AI-generated code in production.
Posted on: Jun 22, 2026 | Read more
Amazon Aurora Serverless – 30% better performance, smarter scaling, scales to zero
Aurora Serverless got a triple upgrade: up to 30% better performance, smarter scaling algorithms that respond to workload changes faster, and it still scales all the way to zero when idle. The “scales to zero” part matters because it means your dev and staging databases genuinely cost nothing overnight, while your production database gets snappier during peak hours. It’s the rare update that simultaneously makes things faster, smarter, and cheaper.
Posted on: Apr 21, 2026 | Read more
AWS Step Functions – AgentCore-powered agentic reasoning step
Step Functions married orchestration with intelligence. The new agentic reasoning step, powered by AgentCore, lets you drop an AI decision-maker right into your state machine. Instead of hardcoding every branch and condition, the agent reasons about what to do next based on the context it receives. It’s like giving your workflow a brain at exactly the points where rigid if/else logic breaks down. Deterministic when you want it, intelligent when you need it.
Posted on: Jun 3, 2026 | Read more
AI & Agents
Amazon Quick – From assistant to autonomous agent platform
Quick went from “helpful chatbot” to “digital workforce” in one spring. The April launch brought a desktop app, Free/Plus pricing (no AWS account needed!), visual content generation, and integrations with Google Workspace, Zoom, Airtable, Dropbox, and Teams. Then in June, the Summit dropped autonomous agents — background workers with specific expertise, tone, and tool access that monitor your CRM, emails, and Slack to proactively act on your behalf. Add a redesigned activity feed that learns your priorities and 16 new integrations (Adobe, Moody’s, Snowflake…), and Quick is no longer just an assistant. It’s an entire team that never takes lunch breaks.
Note for EU-based organizations: The Free and Plus tiers currently process data in a US region. If your compliance requirements mandate EU data residency, the Enterprise license is the only option that supports EU-region processing.
Posted on: Apr 28 & Jun 17, 2026 | Read more
AWS Context – A knowledge graph for your agents
Here’s the problem: your agents are smart, but they don’t know where the bathroom is — metaphorically speaking. AWS Context automatically maps relationships across your existing data into a knowledge graph and makes it available to every agent in your organization. It understands which tables exist, what’s in each column, which sources are authoritative, and how everything relates. As agents interact with it, Context learns which paths produce correct results. It’s like giving your agents a senior employee’s institutional knowledge on day one — except this employee never forgets and never leaves for a competitor.
Coming soon | Read more
Amazon Bedrock AgentCore – The agent apartment complex gets upgrades
AgentCore got a full renovation this spring. Managed Knowledge Base delivers turnkey RAG pipelines with native connectors and an Agentic Retriever for multi-step queries. Web Search lets agents ground responses in current web knowledge without data leaving your environment. And AgentCore Harness hit GA — meaning you can now define an agent’s model, tools, skills, and instructions in config and have it running in minutes, no orchestration code required. It’s gone from “build your own agent plumbing” to “just describe what you want and ship it.”
Posted on: Jun 17, 2026 | Read more
AWS MCP Server – Generally available
The managed Model Context Protocol server graduated from preview and earned its diploma. It gives AI agents and coding assistants secure, IAM-authenticated access to all AWS services through a standard interface. With GA came IAM context keys (no separate permission needed) and unauthenticated documentation retrieval. Your coding agents can now interact with AWS services like a seasoned solutions architect who happens to have perfect memory of every API call ever documented.
Posted on: May 6, 2026 | Read more
Claude Platform on AWS – Anthropic goes native
Anthropic’s full Claude platform is now available directly through your AWS account — authentication, billing, commitment retirement, the works. You get Managed Agents (beta), web search, code execution, Files API, Skills, MCP connector, and the Claude Console for prompt development. It’s like Anthropic opened a branch office inside your AWS account and said “everything we have, you can use here.” Claude stays on Bedrock too, but now you have options depending on whether you want AWS-as-data-processor or the full native platform experience.
Posted on: May 11, 2026 | Read more
OpenAI models on Amazon Bedrock – The partnership deepens
GPT-5.5, GPT-5.4, and Codex landed on Bedrock in limited preview, and Amazon Bedrock Managed Agents powered by OpenAI joined them. The message is clear: AWS wants to be the Switzerland of AI models — neutral ground where you pick the best model for the job without worrying about infrastructure, security, or billing gymnastics. Your existing Bedrock APIs, governance, and cost controls work unchanged. Just swap the model ID and go.
Posted on: Apr 28, 2026 | Read more
xAI Grok 4.3 on Amazon Bedrock – The model marketplace widens
The “Switzerland of AI” strategy got another flag on the map. xAI’s Grok 4.3 arrived on Bedrock with a million-token context window, aggressive pricing, and enterprise-grade reasoning capabilities. Between OpenAI, Anthropic, Amazon’s own Nova models, and now xAI, Bedrock is becoming the one platform where you can genuinely pick the best model for each job and swap between them without changing a single line of infrastructure code. Competition for your inference dollar has never been fiercer — and customers are the ones who win.
Posted on: Jun 20, 2026 | Read more
Amazon Connect expands into four agentic solutions
Connect is no longer just a contact center — it’s now a family of four. Decisions handles supply chain planning with 25+ specialized tools. Talent (preview) runs AI-led interviews with science-backed assessments. Customer (the original Connect) got a conversational AI setup that takes weeks instead of months. And Health delivers patient verification, appointment management, and ambient documentation for healthcare. Amazon basically took 30 years of operational science and packaged it into four agents wearing different hats.
Posted on: Apr 28, 2026 | Read more
Amazon Bedrock AgentCore Payments (Preview) – Agents that transact
Agents that transact. AgentCore now includes a payments capability that lets your agents handle purchase transactions within guardrails you define. Your procurement agent can actually buy the thing it recommends. Your customer support agent can issue a refund without escalating to a human. It’s the moment agents go from “here’s what I suggest” to “done, here’s your receipt” — and it changes the ROI math on every agent you build. Admittedly, this is also a slightly scary feature: autonomous software spending real money on your behalf requires a level of trust that most organizations are still developing. The guardrails had better be good.
Posted on: May 7, 2026 | Read more
Amazon WorkSpaces for AI Agents – Desktop apps meet autonomy
Your AI agents now have their own desktops. WorkSpaces for AI Agents lets your autonomous agents interact with desktop applications — think SAP, legacy ERPs, or anything else that never got an API. Instead of building fragile screen-scraping integrations, your agents now operate applications the same way a human would: clicking buttons, filling forms, reading screens, and navigating menus in an isolated virtual desktop. It’s the bridge between the agentic future and the enterprise software that isn’t going anywhere anytime soon.
Posted on: Jun 30, 2026 | Read more
Security
AWS Continuum – Security at machine speed
In a world where AI finds vulnerabilities faster than humans can patch them, AWS built Continuum — an AI-native security platform that continuously discovers, validates exploitability, prioritizes by business context, and drives remediation through your existing processes. It’s model-agnostic, fully auditable, and every outcome feeds back into the next cycle. Add threat modeling that auto-generates STRIDE analyses from your design docs or source code, and you’ve got a security team member that never sleeps and never says “we’ll get to it next sprint.”
Posted on: Jun 17, 2026 | Read more
AWS WAF AI Traffic Monetization
Plot twist: instead of just blocking AI bots, you can now charge them. AWS WAF’s new Bot Control capability lets content providers set a price for AI bot access, accept payment through third-party providers, and grant scoped access at the edge. It’s like putting a tip jar at your API’s front door — except it’s actually a toll booth with a receipt printer. Publishers who’ve been watching AI scrapers feast on their content for free finally have a billing button.
Posted on: Jun 15, 2026 | Read more
Storage
Amazon S3 Files – Your buckets are now file systems
After 20 years of “S3 is object storage, not a file system,” AWS said “why not both?” S3 Files makes any bucket mountable as a high-performance NFS file system with ~1ms latencies and 25,000 simultaneous connections. No more copying data between S3 and EFS, no more choosing between durability and interactive access. Lambda functions can mount S3 buckets directly. It’s the announcement that finally ends the eternal “should I use S3 or EFS?” debate — the answer is now “yes.”
Posted on: Apr 7, 2026 | Read more
Amazon S3 Annotations – Rich metadata for the agentic era
S3 objects can now carry up to 1 GB of rich, mutable, queryable context directly attached to them. Purpose-built for AI agents and autonomous workflows that need to discover, understand, and act on data at scale. No more maintaining separate metadata systems or external catalogs — the context travels with the data. It’s like giving every S3 object its own LinkedIn profile that agents can browse to understand what they’re looking at before deciding what to do with it.
Posted on: Jun 16, 2026 | Read more
Developer Experience
Kiro for iOS – Your IDE in your pocket
Kiro, the AI development agent, escaped the laptop. The new iOS app lets you kick off projects, monitor progress, steer agents, review diffs, and approve changes from your phone. Sessions run in an always-on cloud environment — start a feature on the subway, review it between meetings, approve during lunch, and continue seamlessly on your laptop. It’s the first time “I’ll work on it during my commute” actually means writing production code without opening a laptop.
Posted on: Jun 17, 2026 | Read more
AWS DevOps Agent – Release Management
Your DevOps Agent now does more than write code — it helps you ship safely. The new release readiness review validates changes against your natural-language standards, and autonomous release testing spins up production-like environments to verify everything works before it touches production. It’s like having a QA team that reads your runbook, understands your deployment policies, and never rubber-stamps a release just because it’s Friday afternoon.
Posted on: Jun 17, 2026 | Read more
AWS Transform – Continuous Modernization
Tech debt finally met its exterminator. Transform’s new continuous analysis (preview) automatically scans your repos against configurable baselines and generates findings in hours, not weeks. Then autonomous remediations create pull requests for affected repositories. It’s like hiring a senior engineer whose only job is to systematically eliminate your accumulated technical shortcuts — except this one works 24/7 and doesn’t judge your past decisions.
Posted on: Jun 17, 2026 | Read more
Operational Excellence
AWS FinOps Agent (Preview)
Your cloud bill now has its own investigator. FinOps Agent automatically detects cost anomalies, traces them to root cause, and delivers explanations straight to Slack or Jira — not another dashboard you’ll forget to check. It runs on schedule, on anomaly events, or on demand. Finally, an answer to “why did our bill spike 40% last Tuesday?” that doesn’t require three engineers and half a day of CloudWatch detective work.
Posted on: Jun 9, 2026 | Read more
Amazon Cognito Multi-Region Replication
Cognito finally learned about geography! Multi-Region replication automatically synchronizes user data, credentials, and pool configurations to a secondary Region, enabling uninterrupted authentication during regional failovers — without forcing password resets. It also adds customer-managed KMS keys for encryption control. Your users won’t even notice when an entire Region has a bad day, and your security team gets the encryption keys they’ve been asking for.
Posted on: Jun 3, 2026 | Read more
Amazon EKS Version Rollback – The undo button for Kubernetes upgrades
Kubernetes upgrades finally have a safety net. EKS Version Rollback lets you reverse a cluster upgrade within seven days if things go sideways — no cluster rebuilds, no data loss, no panic. The control plane reverts to the previous Kubernetes version while preserving all etcd data, workloads, and persistent volumes. It turns what used to be a one-way, hold-your-breath operation into something you can actually try with confidence. For anyone who’s ever stared at a post-upgrade broken cluster at 2 AM, this is the feature you’ve been asking for since EKS launched.
Posted on: Jul 1, 2026 | Read more
Amazon CloudWatch – Native OpenTelemetry metrics with PromQL and per-GB pricing
CloudWatch just tore down its pricing wall. Native OpenTelemetry metrics mean you can send standard OTel data straight in without conversion gymnastics. PromQL querying means your team’s existing Prometheus queries work as-is. And per-GB pricing means you pay for what you actually ingest instead of playing the “how many custom metrics can we afford?” game. For teams that have been agonizing over whether to self-host Prometheus or go all-in on CloudWatch, this might be the answer that ends the debate.
Posted on: Jun 16, 2026 | Read more
The Big Picture
If 2025 was about introducing agents, 2026 is about making them production-ready at enterprise scale. The pattern is clear:
- Run (Lambda MicroVMs, AgentCore Harness, Aurora Serverless) — secure, scalable execution environments
- Know (AWS Context, S3 Annotations, MCP Server) — agents that understand your business
- Act (AgentCore Payments, WorkSpaces for AI Agents, Step Functions reasoning) — agents that transact, operate legacy apps, and make decisions
- Secure (Continuum, WAF monetization) — governance that keeps pace with autonomy
- Ship (DevOps Agent, Transform, Kiro mobile, EKS Rollback) — faster development without cutting corners
- Observe (CloudWatch OTel + PromQL, FinOps Agent) — visibility that scales with complexity
- Choose (OpenAI + Anthropic + xAI + Amazon on Bedrock) — model flexibility without lock-in
The question is no longer “should we use AI agents?” — it’s “how fast can we put them to work?” AWS is betting the answer is “immediately.”
This post was compiled by NordHero — an AWS Advanced Tier Partner helping Nordic companies build and modernize on AWS.