AWS's latest weekly roundup, published July 20, covers a new one-click setup prompt for configuring coding agents with serverless best practices, the arrival of OpenAI's GPT-5.6 model family on Amazon Bedrock, and a handful of smaller but practically useful service updates.
The update that caught the most attention internally was a new agent setup prompt for AWS Lambda. Pasting the prompt's URL into a supported AI coding agent configures it with AWS Serverless skills and the Serverless Model Context Protocol server in one step, embedding serverless best practices into the agent's behavior from the start rather than relying on the developer to know them. The prompt references a Lambda agent setup guide with installation commands for Claude Code, Kiro, Cursor, GitHub Copilot, Codex, Devin Desktop, and OpenCode, and developers can trigger it either from a button on the Lambda console or by copying the URL directly into their agent of choice.
On the model side, Amazon Bedrock added OpenAI's newest model family, spanning three capability tiers built for different tradeoffs between reasoning depth and cost. Sol sits at the flagship reasoning tier, Terra targets balanced performance, and Luna is tuned for fast, cost-efficient inference, with all three accessible through the Responses API on Bedrock's next-generation inference engine.
Amazon S3 picked up more flexibility in how objects move to cheaper storage tiers. Customers can now transition objects into S3 Standard-Infrequent Access or S3 One Zone-Infrequent Access as soon as the day they're created, removing the previous 30-day minimum retention requirement in S3 Standard. Both infrequent-access tiers offer up to 40 percent lower storage costs than S3 Standard while still delivering millisecond retrieval when data is needed, which the roundup notes is well suited to backups, log analytics, and compliance data that goes cold within hours rather than weeks.
AWS Lambda also gained support for self-managed S3 buckets as a code source, letting functions reference source code directly from a customer's own bucket rather than having Lambda create an intermediate copy. That removes prior code storage limits and shortens function activation time after creates and updates by cutting out the copy step entirely.
On the identity side, Amazon Cognito now supports importing users with existing password hashes through CSV user imports. Previously, imported users had to reset their password on first sign-in regardless of what credentials they already had; now administrators can include password hashes in the import and specify the hashing algorithm their source system used, letting migrated users sign in immediately with their existing credentials.
The roundup also marked Amazon SQS's 20th anniversary, noting the service's core decoupling pattern, letting producers and consumers communicate asynchronously without cascading failures, remains the reason customers reach for it two decades after launch, alongside newer posts on combining open agent protocols like MCP and A2A with the Strands Agents SDK, and on using Kiro CLI's MCP integration to speed up AWS support case workflows across scenarios like Glue job failures and Lambda cold-start investigations.
Finally, the update addressed a billing dashboard issue from the prior weekend, in which some customers received erroneous budget and cost anomaly alerts along with inflated estimated cost and usage figures. AWS said the issue has been resolved, all services are operating normally, and a retrospective is underway to prevent similar incidents and improve its response process for future billing issues.