re:Invent 2025 Preview: What We're Watching

re:Invent 2025 starts in a couple of weeks. Rather than predict specific announcements (a lossy exercise), this post is a short list of the areas we’re watching, and the questions we’ll be evaluating each announcement against.

What we expect to see emphasized

Agentic AI for operations. The 2024 keynotes hinted at it. The 2025 keynotes will almost certainly make it concrete. Expect agents that can work across services (investigate Cost Explorer anomalies, modify CloudFormation, respond to GuardDuty findings) and the operational tooling to govern them.

Multi-region resilience. Aurora DSQL has been GA since May; expect post-GA enhancements (broader Postgres feature parity, more region pairs, possibly tooling integrations). Plus supporting announcements: better cross-region networking, more services with native multi-region modes, possibly improvements to Route 53 ARC.

FinOps tooling consolidation. Cost Explorer, Compute Optimizer, Trusted Advisor, and the various per-service cost dashboards continue to fragment the cost-management experience. We expect a real consolidation push, possibly with AI-driven recommendation flows.

Developer productivity for IaC. Q Developer will get more attention. We expect more agentic capabilities for CloudFormation and Terraform authoring, possibly direct integration with CodePipeline review flows.

Questions we’ll be asking

For each major announcement, we’ll be evaluating against:

1. Does it reduce real operational toil for the average team, or only for very large or very specialized ones? Many AWS announcements are for the long tail of advanced users. The most useful announcements lift the floor for everyone.

2. What’s the migration cost vs. status quo? Even good services aren’t worth migrating to if the existing solution is working. We’ll be asking whether each announcement is worth disturbing existing setups for.

3. Is this a managed version of something we currently run ourselves? These are usually the highest-value announcements. S3 Tables in 2024 was a perfect example.

4. Is the pricing actually attractive? Some announcements look great in the keynote and don’t survive the spreadsheet.

What we won’t get excited about

A few categories we’ll cover briefly and move past:

• New Bedrock model announcements without significant capability gains
• Console UI improvements
• Partner integrations that don’t change our customers’ choices
• Generic “AI for X” rebadging of existing services

Stay tuned

We’ll publish a recap in December covering the announcements we think actually matter for AWS DevOps work in 2026. In the meantime, drop us a note if there’s a specific announcement area you’d like our take on.