Wednesday, September 16, 2026

Percona has released version 8.4.11-11, which includes targeted patches designed to improve overall performance. This update builds upon the broader performance progression observed in the 8.4 series released earlier in 2026. The release focuses on specific optimizations to enhance efficiency for MySQL workloads.

• Version 8.4.11-11 introduces patches for significant performance gains.
• Optimizations build on the 8.4 series progression established in 2026.
• Review prior 8.4 performance analysis for full context on these changes.
• Consider testing this patch version to validate efficiency improvements.

PostgreSQL 19 introduces two new contrib modules to stabilize query execution plans when data growth causes the planner to select suboptimal paths. pg_plan_advice allows users to capture a specific plan as a string and enforce it for future executions. pg_stash_advice automates this by storing these plan strings keyed by query ID and applying them automatically.

• Stabilizes queries that degrade after ANALYZE due to data growth without code changes
• pg_plan_advice enables manual capture and enforcement of specific execution plans
• pg_stash_advice automates plan enforcement using query ID keys for seamless operation
• Both modules are part of the contrib extension and available in PostgreSQL 19 Beta 3

Agoda migrated its 1.5 TB hotel Price Cache from a complex 72-shard SQL Server architecture to two DragonflyDB clusters. The team executed the migration using staged dual reads, parity validation, and gradual traffic shifting to ensure data consistency. Post-migration, the organization observed an eightfold reduction in P99 read latency while maintaining high availability.

• Migrating from 72 SQL Server shards to DragonflyDB simplified infrastructure for a 1.5 TB dataset.
• Staged dual reads and parity validation ensured data integrity during the traffic shift.
• P99 read latency improved approximately eightfold after the migration.
• Two DragonflyDB clusters provided the necessary high availability for the price cache.

Amazon SageMaker JumpStart now hosts NVIDIA-optimized Qwen3.6-35B-A3B-NVFP4 and Alibaba's Wan2.1-T2V-1.3B-Diffusers models. The Qwen variant is a Mixture-of-Experts architecture with 35B total parameters but only 3B active per token, featuring NVFP4 quantization for agentic coding and long-context tasks. Wan2.1 provides lightweight text-to-video generation capabilities directly to AWS customers.

• Qwen3.6 uses MoE with 3B active parameters per token, reducing inference compute vs full 35B dense models.
• NVFP4 quantization on Qwen3.6 optimizes memory and speed for agentic coding and multimodal reasoning.
• Wan2.1 offers a 1.3B parameter diffusion model for efficient text-to-video generation workflows.
• Both models are immediately deployable via SageMaker JumpStart for scalable AI solution building.

Mistral AI's Ministral-3-3B and 8B instruct models are now available on Amazon SageMaker JumpStart. The 3B variant is optimized for ultra-lightweight edge deployment, combining a 3.4B language model with a 0.4B vision encoder. It requires only 8GB of VRAM in FP8 format while supporting a 256K-token context window.

• Ministral-3-3B fits in 8GB VRAM (FP8) for resource-constrained edge environments.
• Supports 256K-token context windows despite its compact model size.
• Integrates a 0.4B vision encoder for multimodal understanding capabilities.
• Expanded SageMaker JumpStart portfolio adds purpose-built foundation models.

Amazon SageMaker JumpStart now hosts Google DeepMind's Gemma-4-31B-it-assistant and NVIDIA's Gemma-4-31B-IT-NVFP4 models. These variants of the 31B dense architecture support multimodal reasoning, coding, and agentic workflows with a 256K context window. The availability includes both a standard assistant-tuned version and an NVIDIA-optimized quantized option for scalable deployment.

• SageMaker JumpStart adds Gemma-4-31B in full-precision and NVIDIA NVFP4 quantized forms.
• Models support multimodal inputs including text, images, and video frame sequences.
• Optimized for agentic workflows and coding tasks with a 256K context window.
• Provides enterprise-ready deployment paths for high-performance AI on AWS.
• NVIDIA variant offers quantization benefits for efficient inference scaling.

This paper introduces inference networks, a graph-based framework where nodes represent distinct LLMs and edges define conditional activation paths. The approach addresses the lack of principled strategies for combining expert models in ensemble or series configurations. By dynamically routing queries based on complexity, the system aims to balance inference costs with performance. It provides a structured method to leverage multiple models rather than relying on a single large model for all tasks.

• Uses graph structures to map conditional activation paths between different LLMs.
• Enables dynamic routing of simple queries to cheaper models and complex ones to powerful ones.
• Addresses the gap in systematic strategies for combining expert models in synergy.
• Offers a framework to optimize the trade-off between inference cost and model performance.

ZGCM-1 is a fully open 7B dense foundation model trained from scratch to handle math and agentic search tasks efficiently. It addresses parametric capacity limits by coupling deliberate internal thinking with active external tool use. The training recipe features an end-to-end high-efficiency design, including interleaved gated sliding-window and full attention mechanisms, and stable FP8 Muon optimization.

• Compact models can exceed parametric limits via internal reasoning + external tool use.
• Architecture uses interleaved gated sliding-window and full attention for 256K context.
• Training employs stable FP8 Muon optimizer for system and algorithmic efficiency.
• Progressive curriculum scales context from 16K to 256K during mid-training.
• Interaction traces are reformulated into Markov Decision Processes for better learning.

TimeThink addresses the inability of current timeseries multimodal LLMs to handle out-of-distribution compositional questions. The framework uses synthetic data to elicit explicit reasoning capabilities, moving beyond the implicit answers that fail in high-stakes domains. It specifically targets the gap where reinforcement learning models struggle with dynamic temporal patterns and complex queries.

• Current TS-MLLMs lack explicit reasoning for high-stakes applications like healthcare.
• TimeThink provides a synthetic framework to elicit compositional timeseries reasoning.
• Addresses failures in out-of-distribution queries that plague RL-based models.
• Enables better handling of dynamic temporal patterns through structured primitives.

Amazon SageMaker AI now allows users to submit a prioritized list of instance types for training and processing jobs instead of specifying a single option. This feature automates capacity selection, enabling the service to pick the first available instance from the provided list. It eliminates the need for complex custom retry logic or concurrent job submissions during periods of high GPU demand.

• Submit multiple instance types in a prioritized list to let SageMaker select available capacity automatically.
• Reduces job submission latency during peak demand without requiring custom orchestration scripts.
• Simplifies workflows for workloads that function equivalently across different instance sizes or types.
• Removes the operational overhead of managing concurrent jobs or complex retry mechanisms.

Amazon SageMaker JumpStart now hosts three new foundation models: IBM's granite-speech-4.1-2b, Kakao's kanana-2-30b-a3b-instruct, and OpenFold3. These additions expand AWS's AI portfolio with specialized capabilities for multilingual speech recognition, bilingual agentic workflows, and biomolecular structure prediction. The models are ready for deployment on AWS infrastructure to address specific enterprise AI challenges.

• IBM granite-speech-4.1-2b handles multilingual ASR and translation across six languages.
• Kakao kanana-2-30b-a3b-instruct enables bilingual agentic AI capabilities.
• OpenFold3 provides high-performance biomolecular structure prediction.
• All models are immediately deployable via SageMaker JumpStart.

Researchers propose OrchSLM, a non-interactive orchestration paradigm that combines heterogeneous small language models to handle agentic subtasks. This approach addresses the latency, privacy, and cost constraints of large cloud-based models while mitigating SLM limitations in long-horizon reasoning. The study probes the dynamics of this complementary architecture as an alternative to interactive debate or iterative verification strategies.

• SLMs offer lower latency and better privacy for repetitive agentic subtasks compared to monolithic LLMs.
• OrchSLM uses a non-interactive, heterogeneous model pipeline to overcome SLM context window limits.
• This paradigm avoids the overhead of iterative verification and debate required by smaller models.
• Specialized SLMs may outperform general LLMs on narrowly scoped tasks in agentic workflows.
• The approach aims to reduce computational costs while maintaining reasoning capabilities.

AWS has integrated TwelveLabs Marengo 3.0 into Amazon Bedrock Managed Knowledge Base, enabling direct multimodal embeddings for video, audio, and images. Unlike previous text-only approaches that relied on transcription, Marengo 3.0 encodes visual scenes and speech cues directly into compact 512-dimensional vectors. This allows for natural language search across media assets with state-of-the-art retrieval accuracy and precise segment timing.

• Marengo 3.0 enables direct multimodal embedding of video, audio, and images in Bedrock KB.
• Captures visual and audio semantics beyond what transcription-based text embeddings provide.
• Produces compact 512-dimensional vectors for efficient storage and high-accuracy retrieval.
• Supports natural language search with precise start/end segment results for media assets.

As AI agents execute longer tasks, their execution logs become too large for manual review, making automated root-cause attribution (RCA) essential. Current LLM-based RCA methods struggle with low accuracy because relevant failure signals are sparse and distributed across distant actions in the trace. This research reframes RCA as a continual search problem to better locate these disconnected clues within massive execution histories.

• Long-horizon agent logs are too large for human review, automating RCA is mandatory.
• Standard LLM RCA fails on sparse, distributed signals across long execution traces.
• Reframing RCA as a search problem improves diagnostic accuracy for complex agents.
• One-shot LLM judgments are insufficient for tracing failures in multi-step tasks.

OdoBot is a new web-agent architecture that significantly reduces token processing costs compared to conventional agents. It achieves this efficiency by constructing a behavioral model of the target application derived from successful task-execution demonstrations. Validated on 45 tasks within the Canvas Learning Management System, the approach demonstrates that leveraging application behavior patterns can lower operational expenses.

• Leverage behavioral models from past successes to optimize future agent runs
• Web agents can reduce token overhead by encoding application-specific patterns
• Canvas LMS experiments confirm cost savings across 45 distinct tasks
• Shift from brute-force UI analysis to behavior-guided execution strategies

A new agentic system addresses supply chain analytics complexity by using a coordinator agent to interpret user intent and delegate tasks to specialized sub-agents. This architecture supports both exploratory analysis and deterministic workflows, allowing planners to seamlessly switch between ad hoc queries and structured processes. The framework aims to bridge the gap between business decision-making and the technical expertise required for database querying, KPI analysis, and forecasting.

• Coordinator agent delegates sub-tasks to specialized agents, reducing the need for cross-functional expertise.
• Supports both ad hoc exploratory analysis and deterministic workflows for flexible planning.
• Integrates database querying, KPI analysis, and demand forecasting into a unified agentic interface.
• Aims to lower the barrier for planners to execute complex data engineering and operations research tasks.

AWS has integrated Amazon Q Console with CloudTrail to enable natural language investigation of API activity. Users can query logged events, check trail configurations, and identify logging gaps without writing code or parsing logs manually. This capability supports security audits and operational troubleshooting by allowing direct questioning of data sources and IAM role access.

• Investigate CloudTrail events using plain English questions instead of writing queries.
• Verify trail configurations and identify gaps in logging coverage automatically.
• Troubleshoot operational issues and security concerns without manual log parsing.
• Confirm which data event sources are currently being tracked by the trails.

Jeremy Schneider details the challenges of generating Software Bill of Materials (SBOMs) for PostgreSQL extensions within containers by relying on AI agents. He outlines the specific structure of current CloudNativePG images and discusses maintaining a downstream fork to include Open Source code restricted by CNCF policies. The work involves tracking upstream build infrastructure and submitting patches to minimize merge conflicts while ensuring full inventory and provenance.

• AI agents can struggle with complex SBOM generation for Postgres extension containers.
• Current CloudNativePG images have a specific structure requiring careful inventory tracking.
• A downstream fork allows inclusion of Open Source code banned by CNCF guidelines.
• Patch management focuses on minimizing merge conflicts with upstream build infra.

This paper addresses the 'attestation deficit' where organizations hold governance policies but cannot produce auditable evidence of enforcement quickly enough. It proposes an adaptive intelligence architecture designed to enforce AI policies in real-time. The analysis cites recent reports highlighting widespread gaps in access controls and agent-to-agent monitoring across enterprises.

• Governance lags behind AI adoption due to an inability to generate timely, tamper-evident evidence.
• Proposed architecture aims to close the gap between policy definition and real-time enforcement.
• Empirical data shows 92% of breaches lack proper access controls, underscoring the urgency.
• Only 17% of organizations currently have agent-to-agent monitoring coverage.
• The framework seeks to make governance auditable within regulatory timelines.

AWS DevOps Agent now allows engineers to manage production operations across AWS, multicloud, and on-premises environments directly within Slack. This update consolidates incident investigation and communication into a single thread, eliminating the need to context-switch between platforms. Teams can initiate investigations by @mentioning the agent in private channels, with all findings and actions captured there.

• Consolidates incident response and investigation tools into Slack to reduce context switching.
• Supports operations across AWS, multicloud, and on-premises environments.
• Enables bidirectional communication via @mentions in private Slack channels.
• Captures all agent findings, team context, and recommended actions in a single thread.

AWS now distributes root user sign-in traffic across US East (N. Virginia), US East (Ohio), and US West (Oregon), reducing single-region dependency. The service automatically routes requests to a healthy region without user intervention. CloudTrail ConsoleLogin events for root users are now logged in the specific Region that processed the authentication request.

• Root sign-in is now resilient across three US East and West regions.
• Traffic routing is automatic; no user action is required for sign-in.
• CloudTrail logs root events in the processing region, not just Virginia.
• Update monitoring and alerting to cover all three regions for full visibility.
• This change applies to all AWS accounts and is available immediately.

AWS has introduced a Detected Anomalies widget for Billing and Cost Management Dashboards, unifying cost tracking with anomaly detection. The widget displays the count and total cost impact of anomalies relative to month-to-date spend, while allowing users to drill into root causes and duration. Users can configure the view with 30, 60, or 90-day look-back periods and filter by severity, service, account, and region.

• Centralize cost, usage, and anomaly data in a single dashboard view.
• Filter anomalies by service, account, region, and severity for targeted analysis.
• Review root cause and duration details directly within the widget.
• Select 30, 60, or 90-day look-back periods to assess trend context.

AWS Glue zero-ETL integrations for Salesforce have been updated to include archived records in target tables, ensuring a complete data replica for analytics. The isArchived field accurately reflects the status of records, maintaining synchronization between the analytical store and the Salesforce source. This update automatically applies to new integrations and backfills existing ones without requiring manual intervention.

• Zero-ETL Salesforce jobs now include archived records for complete analytics coverage.
• The isArchived field tracks status, keeping analytical data in sync with Salesforce.
• No action required; new integrations capture archived data automatically from creation.
• AWS Glue automatically backfills archived records for existing integrations.

AWS Glue zero-ETL integrations now track which integration owns the properties of a target table, preventing silent collisions between multiple pipelines. This feature enforces isolation by detecting conflicts when different integrations attempt to write to the same destination catalog. Users are guided to either select a new target or update the existing integration, ensuring predictable data landing across S3 Tables and SageMaker Lakehouse.

• Prevents multiple zero-ETL integrations from silently overwriting the same target table
• Tracks ownership of table properties to the specific integration that created them
• Validates conflicts at creation or modification time for S3 Tables and SageMaker Lakehouse
• Guides users to resolve conflicts by choosing different targets or updating existing configs
• Ensures data pipelines remain isolated and land exactly where intended

Percona and HexaCluster have formed a partnership aimed at reducing the risk and timeline of migrating away from proprietary databases like Oracle, SQL Server, DB2, and Sybase ASE. The collaboration combines Percona’s open source distributions, operators, and enterprise support with HexaCluster’s assessment and migration engineering capabilities. This joint offering is designed to provide a predictable timeline without requiring extensive multi-year consulting engagements.

• Partnership targets migration from Oracle, SQL Server, DB2, and Sybase ASE to open source alternatives.
• HexaCluster provides the assessment and engineering layer for the migration process.
• Percona supplies the underlying open source distributions, operators, and enterprise support.
• Goal is to eliminate multi-year consulting programs in favor of predictable timelines.

DeskcommCRM is a self-hosted, multi-tenant CRM designed to replace platforms like Kommo and Intercom for chat-based sales. It features native AI agents that handle customer interaction and qualification directly within WhatsApp via WAHA. The system is built with data sovereignty in mind, supporting LGPD compliance and offering multi-language capabilities.

• Self-hosted architecture ensures data privacy and eliminates per-seat licensing fees.
• Native WhatsApp integration via WAHA allows AI agents to qualify leads autonomously.
• MCP-ready design facilitates future integration with broader AI agent ecosystems.
• Multi-tenant support enables secure deployment for agencies or large enterprises.
• One-command production setup available through HostGator partner kit.

Homebrew has released BrewUI, an official native macOS application built with Swift 6.0 and SwiftUI. The tool provides a graphical interface for discovering and managing packages while maintaining full transparency over underlying CLI operations. It interacts directly with the brew CLI and Homebrew JSON API, ensuring users see exactly what is happening during installations and updates.

• Native SwiftUI app targets macOS Tahoe 26+ for CLI-averse users
• Maintains transparency by never hiding underlying Homebrew operations
• Uses Swift 6.0 strict concurrency and Swift Package Manager
• Launches brew via /bin/zsh with --no-rcs to disable shell startup files

OpenResearch is a local-first workspace designed to transform CLI coding agents like Claude Code and Cursor into autonomous research agents. It enables users to review literature, develop hypotheses, and generate research artifacts by connecting local LLMs or custom endpoints. The tool provides a local dashboard for managing these automated research tasks on macOS and Linux.

• Wraps coding agents (Claude Code, Cursor) into research workflows for literature review and hypothesis generation.
• Supports local LLM integration via LM Studio, Ollama, or custom endpoints for privacy and control.
• Runs locally with a web dashboard at 127.0.0.1:4791, requiring no cloud dependency for core features.
• Simplifies setup via a single CLI installer script for macOS and Linux environments.