#agentic-ai

The new database world according to Google: Inexact queries and AI in everything

Google giới thiệu tầm nhìn tương lai về cơ sở dữ liệu với sự tích hợp AI agent và truy vấn không chính xác (inexact queries) thông qua ngôn ngữ tự nhiên, thay vì chỉ SQL truyền thống. Họ đang nhúng các chức năng AI (như AI.IF sử dụng Gemini) trực tiếp vào Google SQL trên các nền tảng Spanner, AlloyDB và BigQuery, đồng thời phát triển "proxy models" để giảm chi phí và độ trễ khi gọi LLM. Trong 3-5 năm tới, người dùng có thể chuyển từ tương tác trực tiếp sang điều phối các agent, thay thế dashboard truyền thống bằng phân tích hội thoại.

Những công nghệ mới từ Google như proxy models và AI tích hợp trực tiếp vào cơ sở dữ liệu sẽ giúp tối ưu hóa hiệu suất và chi phí khi xử lý các truy vấn phức tạp, đồng thời mở ra khả năng tương tác tự nhiên thông qua ngôn ngữ, giúp lập trình viên tìm hiểu cách ứng dụng AI vào hệ thống dữ liệu hiện đại.

How KRAFTON Built PUBG Ally, a Co-Playable Character Powered by NVIDIA ACE

KRAFTON built PUBG Ally, an AI teammate for PUBG: BATTLEGROUNDS powered by NVIDIA ACE, introducing a new category called co-playable character (CPC). The system runs entirely on-device using a pipeline of ASR, a quantized 2B-parameter Mistral-NeMo-Minitron SLM, and custom TTS. Key engineering decisions include a System 1/System 2 architecture separating fast behavior-tree reactions from slower language reasoning, KV cache optimization for low latency, agentic tool calls for grounded game-state awareness, and knowledge distillation from a teacher model for PUBG-specific domain adaptation. The model supports English, Korean, and Chinese, runs on GPUs with as little as 8GB VRAM, and includes cross-session long-term memory. The team iterated through large-scale playtests with over a thousand players to tune the non-deterministic AI outputs.

Two Six Technologies Unveils Sentr, the Command Center for the Information Environment

Two Six Technologies has launched Sentr, an AI-powered command center platform for information operations targeting U.S. government and allied partners. Built on Two Six's Helix agentic orchestration layer, Sentr unifies collection, research, planning, and messaging into a single chat interface. It uses AI agents to process large datasets, identify narrative opportunities, generate action plans, and deploy targeted content globally. Early users report 85% time savings. The platform is already deployed with Department of Defense, Department of State, and Intelligence Community customers, and includes audit trails, doctrine-based AI output scoring, and secure LLM interactions.

The CPU's growing role in agentic AI infrastructure

Arm-sponsored content arguing that CPUs play a critical but underappreciated role in agentic AI infrastructure. While accelerators handle model performance, CPUs act as the control plane — managing data movement, workload scheduling, and secure isolation. Arm's Neoverse platform underpins custom silicon from AWS (Graviton), Google (Axion), Microsoft (Azure Cobalt), and NVIDIA (Grace Hopper/Blackwell), all reflecting a shift toward purpose-built Arm-based processors in cloud and AI datacenters. The piece introduces the Arm AGI CPU, built with Meta, targeting rack-level density for agentic AI deployments.

How to integrate OpenShift AI and PG Airman MCP Server

A deep-dive into the Data Governance Copilot architecture (Part 3), covering container-level design with SvelteKit frontend, FastAPI backend, and an agentic loop integrating OpenShift AI with the PG Airman MCP server. Explains two deployment modes: Red Hat Integrated Llama Stack versus lower-level MCP-direct using the OpenAI SDK and MCP Python library. Details how inbound LLM messages use OpenAI API format converted via Jinja tokenizer templates in vLLM, and how outbound tool calls are parsed — including a custom client-side parser needed for NVIDIA Nemotron Nano 9B due to its lack of streaming support in the native vLLM plugin. Compares Nemotron's XML-tagged TOOLCALL format against Qwen3's standard hermes parser output.

What Is a Multi-Agent System?

Hệ thống đa tác nhân (MAS) gồm nhiều AI agent tự chủ phối hợp giải quyết nhiệm vụ phức tạp, với các thành phần chính như worker agents, orchestrator agents, môi trường thực thi, bộ nhớ chia sẻ, giao thức (MCP, A2A) và chính sách quản trị. Các ứng dụng thực tế bao gồm quản lý hạ tầng của NTT Data, nền tảng nghiên cứu dược phẩm của Madrigal Pharmaceuticals (dựa trên LangChain/LangSmith) và hệ thống chăm sóc sức khỏe của Fujitsu. Gartner ghi nhận sự quan tâm từ doanh nghiệp tăng 1.445% nhờ khả năng tự động hóa quy trình phức tạp ở quy mô lớn.

Lập trình viên nên đọc bài này để hiểu cách xây dựng hệ thống tự động hóa công việc phức tạp bằng cách kết hợp nhiều agent độc lập, từ đó tối ưu hóa hiệu suất và linh hoạt cho các ứng dụng doanh nghiệp tương lai.

More Malicious OpenClaw Skills Threaten AI Supply Chain

Unit 42 researchers at Palo Alto Networks identified five malicious skills on ClawHub, the marketplace for the OpenClaw AI agent framework, that bypassed automated security scanning including VirusTotal and ClawScan. The skills fell into three threat categories: infostealers targeting macOS with C2 connectivity, detection evasion via inflated file sizes exceeding scanner thresholds, and novel agentic threats including affiliate link injection and a cryptocurrency pump-and-dump scheme. This follows earlier reports of widespread malicious skills on the platform, including Bitdefender's finding that 17% of analyzed skills carried malicious payloads. All five skills were reported and removed. Defenders are advised to treat skills as potential network entry points, monitor outbound agent traffic, validate publisher provenance, and apply least-privilege access to agents.

Databricks positioned highest in execution and furthest in vision for the second consecutive year in Gartner Magic Quadrant

Databricks announces it has been named a Leader in the 2026 Gartner Magic Quadrant for AI Platforms for Data Science and Machine Learning, holding the highest position in Ability to Execute and furthest in Completeness of Vision for the second consecutive year. The post highlights Databricks' unified platform philosophy combining lakehouse, Lakebase, Agent Bricks, and Unity Catalog to deliver governed, production-grade agentic applications. Key capabilities include centralized governance via Unity AI Gateway, support for frontier and open-source models, and tools for both developers and business users to build AI agents grounded in enterprise data.

Build Trust in Agentic AI: From POC to Production

A four-step framework for moving agentic AI systems from proof-of-concept to production-grade enterprise deployments. The framework covers: (1) a foundation layer using RAG, operational data, and agent memory to reduce hallucinations; (2) a verification layer with Agent Confidence Score (ACS) and Business Risk Score (BRS); (3) a governance layer using an Agent Decision Score formula (ADS = ACS × (1 - BRS)) with a traffic-light routing system for human-in-the-loop escalation; and (4) an outcomes layer with executive dashboards tracking ROI, cost-per-task, and systemic risk. A customer refund workflow is used throughout as a concrete example. MongoDB is positioned as the unified data platform for agent memory, vector search, and execution traces.

Log management for AI workloads: How to bring your logs and telemetry plan into the AI-first century

AI workloads are overwhelming traditional log management systems, exposing limits in how teams capture, store, and correlate telemetry. A five-point action plan covers: unifying all telemetry signals on a single platform, correlating logs with traces for causation, controlling costs by eliminating rigid schemas and rehydration overhead, standardizing instrumentation at ingest, and enabling preventive operations through real-time contextual analytics. Key stats from a 2026 survey of 450 IT leaders highlight that 85% of organizations struggle to ingest logs at AI scale, 74% cite indexing and rehydration costs as barriers, and 84% say customer trust in AI depends on predictive log analytics.

Beyond regex: Harvesting security logic with LLMs

Dự án SastAI của Red Hat kết hợp AI thế hệ mới (LLMs) với kỹ thuật RAG, semantic code search và pattern harvesting để tự động xác định các cảnh báo sai (false positive) trong SAST. Bằng cách trích xuất logic bảo mật lặp lại từ 8.000 phát hiện đã được chú thích, hệ thống đạt hiệu suất cao hơn 10% so với mô hình gốc, đồng thời hướng tới tích hợp học liên tục và có thể mã nguồn mở.

Lập trình viên nên đọc bài này để khám phá cách AI sinh thành có thể tự động hóa và nâng cao hiệu quả phát hiện lỗ hổng an ninh trong mã nguồn, giúp tiết kiệm thời gian và giảm sai sót trong kiểm thử an ninh ứng dụng.

Enterprise AI’s Last Mile: Scaling ROI with Governed Apps

Enterprise organizations are struggling to move AI projects beyond the proof-of-concept stage into scalable, production-ready systems. The core problem is that AI adoption is outpacing governance, with only 21% of companies having operationalized the foundations needed to scale safely. The proposed solution is building AI-driven applications with governance baked in by design rather than added as an afterthought. BARC research supports this: organizations with company-wide governed data products are far more likely to have three or more AI projects in production (85% vs. 25%). A practical framework is outlined covering seamless application development, integrated governance with automated monitoring, and a shift from experimental pilots to product-oriented AI development.

Rethinking cloud operations with agentic observability

Microsoft has announced the general availability of the Azure Copilot Observability Agent, built on Azure Monitor. The tool correlates logs, metrics, traces, topology, and operational context across environments to help operators move faster from incident detection to root cause resolution. It addresses growing cloud complexity — with 84% of organizations reporting increased complexity and 69% saying it outpaces their operating model — by using AI agents to reason across signals in real time. Customer examples cite reclaiming 250 engineering hours monthly and significantly faster incident resolution. The post frames observability as foundational to a broader shift toward agentic cloud operations, where AI agents continuously interpret signals, take action, and improve system resilience over time, with governance and human oversight remaining essential.

Taming the AI Double Threat with Isovalent Runtime Security

AI is creating a dual security challenge: AI workloads (especially agentic ones) introduce new attack surfaces, while adversaries use AI to compress attack timelines. Agentic workloads are non-deterministic at the application layer but exhibit stable behavioral profiles at the kernel level, making eBPF-based runtime security effective. Isovalent Runtime Security (built on Tetragon and eBPF) addresses both threats through kernel-level enforcement, process/network/file visibility, sandbox policies with default-deny posture, and L7 network introspection. The post argues that existing cloud-native runtime security primitives, extended with the right mental models, are sufficient to secure AI workloads and respond to AI-accelerated attacks.

AWS hypes continuous agentic DevOps, puts Kiro in your pocket

AWS announced several new and enhanced AI-driven DevOps and security tools at its New York Summit. AWS Continuum, now in closed preview, provides continuous vulnerability scanning and penetration testing for AWS environments, with AI-generated fix suggestions. The DevOps Agent gains release management capabilities and now exposes its own MCP endpoint plus Agent2Agent (A2A) protocol support, enabling integration with observability tools like Datadog, Dynatrace, and New Relic. Kiro, the specification-driven AI coding IDE, is getting a native iOS mobile app for managing remote sessions. AWS Transform adds continuous modernization for Java and .NET workloads. Amazon Bedrock AgentCore gains managed knowledge bases, web search, and policy/guardrail features to improve agent reliability and trust. Pricing for these agentic services is noted as opaque and potentially costly, with per-second usage billing and additional charges for consumed AWS services.

What AI Readiness Really Means and How It Applies to APIs

AI readiness goes beyond choosing smart models — it requires mature data, governance, and integration foundations. For APIs specifically, this means shifting from human-readable interfaces to machine-consumable ones that AI agents can reliably discover, call, and recover from. Traditional APIs fail in agentic systems due to poor documentation, inconsistent error handling, non-idempotent behavior, human-driven authentication, oversized payloads, and infrastructure fragility. To become AI-ready, APIs should adopt strongly typed OpenAPI schemas, concrete examples, structured JSON error objects, idempotent endpoints, strict pagination, and machine-friendly auth like OAuth 2.0 client credentials. These upgrades enable enterprises to scale agentic architectures safely without risking compliance failures, cost overruns, or security exposures.