AWS June 8 Roundup: IoT SDK for Swift, Cognito Multi-Region, and AgentCore in Step Functions
The AWS weekly roundup for June 8, 2026 delivers several developer-facing releases across infrastructure, identity, and AI orchestration. The most notable SDK release is the AWS IoT Device SDK for Swift reaching general availability, bringing production-ready MQTT 5 connectivity, Device Shadow, Jobs, and fleet provisioning to Swift developers across macOS, iOS, tvOS, and Linux. This expands the Swift ecosystem into IoT and edge device applications, an area previously requiring either platform-specific wrappers or abandoning the Swift toolchain entirely.
Amazon Cognito gains multi-region replication, allowing user and machine identity data to be synchronised to a standby regional user pool in near real-time. The feature is available across 16 regions, and users remain authenticated during a regional failover without requiring re-authentication — a significant operational improvement for applications serving users who expect uninterrupted session continuity. Amazon RDS for SQL Server adds support for Bring Your Own Media (BYOM) via License Mobility, letting on-premises SQL Server customers reuse existing Microsoft licenses and Software Assurance entitlements when migrating workloads to AWS, with AWS License Manager handling compliance tracking.
On the AI orchestration side, AWS Step Functions introduces AgentCore-powered agentic reasoning steps. This allows developers to embed AI agent reasoning directly into state machine workflows — running multiple agents in parallel or in sequence, adding human approval steps between them, and tracing every decision through standard CloudTrail and CloudWatch integrations. Amazon EKS adds Kubernetes 1.36 support and ECS gains Trainium and Inferentia accelerator support, relevant for teams running distributed ML inference workloads at scale. Amazon Bedrock's redesigned console introduces a unified model catalog, side-by-side model comparison, and project-aware documentation with pre-filled API code snippets, reducing friction when evaluating and switching between foundation models.
Read more — Amazon Web Services
Google Cloud Gemini Enterprise Agent Platform and Cloud Run Worker Pools GA
Google Cloud launched the Gemini Enterprise Agent Platform (GEAP) at Cloud Next '26 as a comprehensive foundation for building, governing, and scaling enterprise AI agents. At its core is the Agent Development Kit (ADK), a modular, model-agnostic framework that treats agent development like conventional software engineering — supporting structured agent graphs, tool calling, memory integration, and evaluation workflows. GEAP is explicitly designed to take agents from experimental environments to production without requiring a rebuild, integrating with Cloud Run for lifecycle management and Cloud Logging and Cloud Trace for observability.
Cloud Run Worker Pools reached general availability alongside GEAP as a new resource type built specifically for pull-based, non-HTTP workloads. Unlike standard Cloud Run services, Worker Pools maintain always-on environments, which avoids cold start penalties for background agent processes that must respond to queue events with low latency. Developers can also now provision individual Cloud Run instances as underlying primitives — paired with Cloud Storage volume mounts — making them suitable for hosting long-running background agents that maintain local state between processing steps. This architecture is particularly well suited to agent loops that consume tasks from a queue, process them over extended periods, and write outputs to structured storage.
Anthropic's Claude Opus 4.8 is now available directly on GEAP for handling complex, multi-stage enterprise workflows, joining Google's own Gemini 3.5 Pro and Flash models in the platform's model catalogue. Google also introduced the Axion N4A Arm-based CPU at Next '26, delivering up to 2x better price-performance than comparable x86 VMs for cost-sensitive workloads such as vector search, embedding generation, and inference serving. The Google Cloud Workbench Notebooks IDE extension is also available, allowing data scientists to connect to and execute notebooks on managed cloud environments from their local IDE without managing remote kernel connections manually.
Read more — Google Cloud
