OffNet Newsroom The Brief

Tuesday, July 21, 2026

6 stories worth your scroll — picked from 50 curated today.

Share on LinkedIn

1 · Copy  →  2 · Share  →  3 · Paste & post

View the post text

LLMs NEW AWS Database Blog

SQL Server 2025 on RDS calls Bedrock via T-SQL

Amazon RDS for SQL Server 2025 enables direct invocation of Amazon Bedrock foundation models from T-SQL using sp_invoke_external_rest_endpoint. This architecture eliminates middleware layers, thereby reducing latency and embedding AI capabilities directly into database workflows. The method allows for agentic AI patterns without requiring external application code to handle model interactions.

Use sp_invoke_external_rest_endpoint to call Bedrock models directly from T-SQL.

Database Technology NEW AWS Database Blog

SQL Server 2025 RDS now calls AWS APIs directly via T-SQL

Amazon RDS for SQL Server 2025 introduces sp_invoke_external_rest_endpoint, allowing direct invocation of AWS services and HTTPS endpoints from within T-SQL scripts. This capability eliminates the need for intermediate application layers to trigger external actions during database transactions. The feature simplifies integrations by enabling server-side logic to interact seamlessly with the AWS ecosystem.

Use sp_invoke_external_rest_endpoint to call AWS or HTTPS endpoints directly from T-SQL.

Agentic AI NEW arXiv cs.AI

Masked Diffusion Models Offer Steerable Text-Based World Models for Agentic RL

Researchers introduce masked diffusion language models as a solution to the left-to-right bias inherent in autoregressive world models. This approach enables better conditioning on globally interdependent state anchors like tool schemas and expected outcomes. The result is a steerable text-based world model that supports diverse, on-demand training environments for reinforcement learning agents.

Addresses mode collapse in RL caused by sparse rewards and fixed task difficulties.

AWS NEW InfoQ

AWS Releases Loom: Open-Source Platform for Governing AI Agents at Scale

AWS has published Loom as an open-source reference platform designed to govern AI agents across enterprise environments. Built on Strands Agents and Bedrock AgentCore Runtime, the implementation enforces security through RFC 8693 token exchange for identity propagation within delegated actor chains. The platform supports config-driven deployments that avoid runtime code generation and mandates strict tagging for resource management.

Loom is a reference implementation, not a managed service, for governing AI agent fleets.

AI / ML NEW arXiv cs.CL

OpenLanguageModel: Readable, Composable SLM Pretraining in PyTorch

OpenLanguageModel (OLM) is an open-source PyTorch library designed for transparent pretraining of small language models. It structures model code to mirror architecture diagrams using explicit wiring modules like Block and Parallel, enabling seamless transitions from educational notebooks to full-scale research ablations. The library integrates tokenizers, streaming datasets, optimization, mixed precision, and hardware-aware execution across CPU and single-node GPU setups.

Code readability mirrors architecture diagrams, easing debugging and teaching.

Oracle Ecosystem NEW InfoQ

Java Roundup: Value Objects Preview, WildFly 41 GA, Oracle AI Agent Studio

The July 13, 2026 Java news cycle highlights the reintroduction of Value Objects in preview, alongside the general availability of WildFly 41. Key updates include Open Liberty 26.0.0.7, maintenance releases for Micronaut and LangChain4j, and new tools like TornadoVM and the Quarkus Shim extension. Oracle also launched an AI Agent Studio specifically designed for Fusion Applications.

Value Objects return as a preview, signaling renewed interest in immutable data modeling in Java.