What AWS’s Latest AI, Database and Cloud Updates Mean for Production Systems

What Happened

AWS announced a set of releases on November 2–3, 2026, spanning AI tools, databases, event processing, security and cost management.

• AI applications: Claude Haiku 5.5 is available in the Kiro IDE and CLI in AWS GovCloud (US), with a 1-million-token context window and five adaptive-thinking effort levels. Users must update and restart Kiro to select it. Amazon Connect Customer flows can now call MCP tools through an External Tool block backed by Bedrock AgentCore Gateway, and its Wait block can resume on External Tool or Cases events. [2][8][9]
• Databases and events: Aurora MySQL 8.4.10 adds GTID tags and supports up to 50,000 simultaneous connections per instance. Aurora MySQL 3.14, compatible with MySQL 8.0.46, adds post-quantum TLS key exchange. Lambda queue mode for self-managed Kafka event sources can increase consumption parallelism, but does not preserve partition ordering and requires Kafka share groups. [5][7][6]
• Operations and security: RDS now reports costs per snapshot by default. Amazon Inspector can scan EC2 AMIs without launching them. CloudFront exposes 28 quotas through AWS Service Quotas, including self-service increase requests. [3][14][15]
• Regional availability: RDS for Oracle added M8i and R8i options in New Zealand and Calgary, with R8i also expanding to Milan, Stockholm, Zurich and Tel Aviv. Textract launched in Tokyo and Zurich, but several specialized document APIs are not available there. [1][13]
• Document processing and access: An automatic Textract text-detection update improves handwriting, long-word and multilingual-table handling without an API schema change. Amazon Quick’s desktop app can now use an organization’s existing Quick web sign-in after an administrator adds the desktop connector. [16][4]

Why It Matters to Businesses

The immediate opportunities are narrower than a blanket upgrade recommendation. Per-snapshot billing makes backup costs attributable; AMI scanning creates a control before image deployment; and Connect’s tool integration can remove custom Lambda glue from some customer workflows. Each changes an operational decision, not just a feature list. [3][14][8]

Kimbodo Engineering Perspective

We would prioritize changes against existing bottlenecks. Lambda queue mode is useful when partition count or slow records constrain throughput, but it is the wrong choice for consumers that depend on partition order. Aurora upgrades warrant application and replication tests even when performed in place. Textract’s automatic model update deserves output-quality monitoring because behavior can change without a schema change. [6][5][7][16]

How We Would Implement It

• Inventory Aurora versions, replication dependencies and connection peaks; test upgrades on restored snapshots, then phase production rollouts through maintenance windows or an upgrade policy. [5][7]
• For Kafka workloads, verify share-group support, test duplicate and out-of-order handling, and compare throughput and cost before enabling queue mode. [6]
• Put AMI scans in image-build pipelines, route findings to Security Hub, and block distribution when images fail the organization’s security standard. Add snapshot cost tags and CloudFront quota alarms to the same operational review. [14][3][15]
• For Connect tools, define narrowly scoped actions and test timeout, failure and Wait-block behavior before exposing them in live flows. Confirm each required feature and API is available in the deployment Region. [8][9][13]

Risks, Costs and Security

Haiku 5.5 has a 0.06x Kiro credit multiplier, rising to 0.3x for requests over 100,000 tokens. Lambda queue mode retains standard Lambda pricing and adds Event Poller Unit charges in provisioned mode. R8i and M8i’s stated price-performance gains are “up to” figures, so workload benchmarks remain necessary. [2][6][1]

MCP tools can retrieve data or trigger third-party actions from a customer flow; permissions, auditability and data exposure need review before launch. AMI scanning helps identify vulnerable software but does not replace runtime controls. Regional feature gaps—particularly Textract’s specialized APIs in Tokyo and Zurich—should be checked before committing to a data-residency design. [8][14][13]

Sources

1. [1] Amazon RDS for Oracle now supports M8i and R8i instances in additional AWS Regions
2. [2] Haiku 5.5 now available on Kiro in AWS GovCloud (US)
3. [3] Amazon RDS now supports per snapshot billing
4. [4] Amazon Quick on desktop now uses your organization's existing Amazon Quick sign-in
5. [5] Amazon Aurora MySQL 8.4.10 is now generally available
6. [6] AWS Lambda now supports queues for Apache Kafka in self-managed Kafka event sources
7. [7] Amazon Aurora MySQL 3.14 (compatible with MySQL 8.0.46) is generally available
8. [8] Amazon Connect Customer now supports Model Context Protocol (MCP) tools in flows
9. [9] Amazon Connect Customer now supports Cases and External Tool events in the Wait block
10. [13] Amazon Textract now available in Asia Pacific (Tokyo) and Europe (Zurich)
11. [14] EC2 Amazon Machine Images can now be scanned for vulnerabilities through Amazon Inspector
12. [15] Amazon CloudFront now supports self-service increases in AWS Service Quotas
13. [16] Announcing an updated text detection model for Amazon Textract