AWS Ships Lambda S3 Files, Bedrock AgentCore CLI, and Aurora Serverless v4 in a Single Week

In its April 27 weekly roundup, AWS announced three features that each would have warranted a standalone post on a slower news week. Together they represent one of the denser single-week releases the platform has shipped in 2026 — all of them aimed squarely at the infrastructure friction points of building and running AI workloads.

Lambda S3 Files: shared state across functions

Lambda functions have long had access to ephemeral /tmp storage (now up to 10 GB) and EFS mounts for shared state. Lambda S3 Files extends that model by letting you mount S3 buckets directly as file systems inside Lambda execution environments. Multiple functions can read and write the same S3 mount simultaneously, without copying data between services.

The practical use case is AI and ML pipelines: multiple processing steps that need to share model weights, checkpoints, or intermediate outputs can now do so through a shared mount instead of passing data through SQS or EventBridge. It works across all Lambda-supported runtimes. Billing uses standard S3 read/write request pricing — no new pricing tier, no EFS configuration required.

Bedrock AgentCore CLI: one command to deploy an agent

Amazon Bedrock AgentCore has been building out a managed runtime for production AI agents since late 2025. The new CLI reduces the deployment surface to three inputs: a model ID, a system prompt, and a tool list. From those three, AgentCore provisions infrastructure across 14 AWS regions at no additional cost.

For teams already running infrastructure as code, CDK support is live now and Terraform integration is on the roadmap. When you’re ready to move beyond the managed runtime, AgentCore can export your agent definition as Strands-based Python, giving you full ownership without rewriting from scratch. That escape hatch matters — managed runtimes that lock you in tend to get abandoned.

Aurora Serverless v4: 30% faster burst scaling

Aurora Serverless v4 is generally available with a 30% performance improvement over v3, driven by a redesigned burst-scaling algorithm. The scale-to-zero behavior is unchanged, keeping it the right choice for development databases and workloads with unpredictable traffic spikes. Both PostgreSQL and MySQL-compatible engines are supported.

The bigger picture

The same week, Meta signed an agreement to run tens of millions of Graviton cores on AWS for CPU-intensive AI inference workloads — a signal that even companies building their own AI infrastructure are hedging on AWS capacity.

Lambda S3 Files and AgentCore CLI are particularly well-timed. Agentic AI workflows stall on shared state management more than almost any other problem: how do you pass context between planning steps, tool calls, and execution loops without stitching together five services? Both features directly reduce that friction. Aurora Serverless v4 handles the persistence layer that agents inevitably need once they move beyond stateless interactions.

AWS is clearly positioning itself as the default compute layer for both AI models (Bedrock) and the agents built on top of them (AgentCore). This week’s releases make that pitch more credible.