Agentic AI

Google's Interactions API is now the default for Gemini models and agents, offering a unified surface for inference, autonomous tasks, and background execution. New features include managed sandboxes, cost tiers, and a migration path from the legacy API.

Ampersend integrated Amazon Bedrock AgentCore Payments to create a pay-per-intelligence layer for AI agents. The two-hop payment architecture enables autonomous, governed transactions without custom billing infrastructure.

Tencent is testing Xiaowei, an AI assistant inside WeChat that uses mini-programs to handle tasks like messaging and ride-hailing. The move intensifies China's AI super-app battle as the company eyes a Q3 2026 launch.

Apple supplier Lingyi iTech is raising up to US$1.1 billion in a Hong Kong IPO to expand into AI hardware and humanoid robotics. The move signals a strategic pivot from smartphones to embodied intelligence, with bold production targets and key partnerships.

A quiet shift is reshaping AI workflows. Instead of typing prompts, developers now design autonomous loops that guide AI agents to complete tasks without human input. This report explores how loop engineering works, its components, and the challenges ahead.

China’s GrainCore Dynamics sees an aging farming workforce as an opportunity to lead in embodied AI for agriculture. Drawing parallels to the EV boom, the company believes its unique blend of AI and mass production can disrupt global agritech.

Tool calling is the mechanism that lets LLMs decide when to invoke external functions, turning them from passive text generators into active agents. This article explains the tool calling loop, how models choose tools, and why it's the foundation of agentic AI.

After a brief frenzy of maximizing AI token use, tech companies are now imposing limits to control costs. The shift from 'tokenmaxxing' to 'tokenminning' reflects a search for real ROI.

Google DeepMind has unveiled an AI control roadmap that redefines agent security. The framework borrows from cybersecurity to treat advanced AI as potential insider threats, adding layers beyond traditional alignment.

French startup Genesis AI, backed by Eric Schmidt and $105M, launches Eno—a non-humanoid robot with human-like hands. It targets factories, labs, and eventually homes, challenging the humanoid trend with a focus on dexterity and practicality.

Pinterest unveils a Business Assistant in Ads Manager, a Model Context Protocol for agentic AI, dynamic creative selection, and the Ask Pinterest experimental app. These tools target its 631 million users, most with shopping intent, to sharpen ad performance and discovery.

China's Ministry of Commerce and seven other ministries released a plan with 17 measures to integrate AI into consumer goods and services, aiming to boost growth through robots, subsidies, and infrastructure. The initiative targets millions of households and shops, focusing on humanoid robots and AI assistants.

Cloudflare introduces Temporary Accounts for Agents, removing sign-up barriers so AI coding agents can deploy Workers directly. The feature uses a new Wrangler flag for instant, autonomous deployments within a 60-minute window.

Union Home Minister Amit Shah announced a three-year target for criminal justice delivery and launched AI-powered tools for predictive policing. The initiative aims to transform India's law enforcement with data analytics and forensic science.

NHAI has launched an in-house AI system that lets officials upload photos to detect road defects and scan project reports for errors. The tool, including the Margsarthi chat agent, has already handled over 50,000 queries.

AWS now offers Mistral AI's Ministral-3-14B-Instruct-2512 in SageMaker JumpStart. The 14B-parameter model combines vision, text, and agentic features like function calling for edge deployment.

Google is releasing Ask Ad Manager, an AI agent that automates routine advertising tasks for publishers and agencies. The beta launches this month, with new skills and developer tools planned throughout the year.

Google's new Home Speaker integrates Gemini AI for multi-step voice commands and sustainable design. Available in four colors, it features a privacy mute switch and light ring underglow.

The U.S. government confirmed Grok AI was used in strikes against Iran, disclosed in a court filing. The revelation exposes the Pentagon's deep reliance on Musk's AI for lethal targeting.

Ukraine has introduced AI-powered interceptor drones like the P1-Sun Long to cheaply counter Russian Shahed swarms. The systems use machine vision trained on combat footage but require human approval for strikes, balancing innovation with ethical oversight.

Amazon S3 introduces annotations, a new metadata capability that attaches up to 1GB of business context directly to objects. This gives AI agents and analytics tools the context they need to discover and understand data without separate metadata systems.

Amazon Bedrock Guardrails now offers a resourceless API for agentic AI workflows, providing per-request control over content filters, prompt attack detection, and sensitive data checks with numeric severity scores. The API enables dynamic, step-level safety without managing guardrail resources.

A new guide details how to run a high-performance local LLM on a Mac Mini for OpenClaw agents, removing recurring API fees. It uses quantization and llama.cpp to achieve practical speeds for everyday tasks.

A new Doctorate in Business AI is drawing C-suite leaders to master AI-native strategy. The program challenges AI-first thinking and prepares executives for flatter, agent-driven organizations.

SpaceX announced a $60 billion deal to buy Anysphere, creator of the AI coding agent Cursor. The move strengthens SpaceX's enterprise AI position and gives Cursor vast computing power.

Casey Harrell, paralyzed by ALS, becomes the first intensive user of a speech brain-computer interface. Meanwhile, South Korea’s AI fervor contrasts with US skepticism, and Anthropic faces export curbs over intelligence risks.

When a production AI agent hits a rate limit, swapping models without adapting the payload can silently corrupt output. Emmanuel I. Michael discovered this failure mode and built a zero-dependency recovery layer that snapshots state, normalizes payloads, and injects resume context to maintain schema integrity during failovers.

Alipay has launched Abao, an AI-powered assistant that transforms its super app into a conversational interface for everyday tasks. The move puts Alipay ahead of WeChat in the race to integrate AI agents into China's most popular platforms.

Strands Evals SDK introduces detectors that automatically identify failures in AI agent execution traces and perform root cause analysis. This slashes manual diagnosis time and provides actionable fix recommendations, integrating seamlessly into evaluation pipelines.

Microsoft’s Build 2026 unveiled Project Solara, a platform for AI agent devices running a secure Android fork. The concept hardware and new MAI models signal a shift from apps to autonomous agents, with enterprise pilots on the horizon.