#AI agents

Alibaba has introduced Qwen3.7-Plus, a multimodal model built to see, click, code, and operate software inside a single agent loop, signaling a sharper push toward autonomous AI systems.

At Build, Microsoft paired OpenClaw enthusiasm with guardrails, using sandboxing and new Autopilots to make AI agents more usable inside companies.

Canva has launched a connector for Perplexity Computer, allowing subscribers to turn prompts and data into editable Canva assets from Perplexity’s desktop agent.

A new review argues that autonomous AI depends less on raw model capability than on the software harness that gives models tools, memory, testing, and execution loops.

Mistral has rebranded Le Chat as Vibe, bundling chat, workplace integrations, coding agents, and recurring workflows into a broader product aimed at competing with enterprise AI agent platforms.

Google Cloud has introduced AI Threat Defense, a platform that combines Gemini, Wiz, DeepMind’s Codemender, and Mandiant tools to find exploitable flaws and generate tested fixes faster.

Robinhood is rolling out a feature that lets customers connect AI agents to a separate investment account through MCP, allowing those agents to buy and sell stocks while regulators warn about oversight and control.

Google used its I/O Dialogues stage to frame AI not as a single product wave but as a platform touching agents, science, robotics, quantum computing and filmmaking.

Google says AI Mode now exceeds one billion monthly users and is rolling out a redesigned AI-powered Search box, Gemini 3.5 Flash as the default model, and new agent-style capabilities.

Google has introduced Gemini 3.5, starting with Gemini 3.5 Flash, a model family built around long-horizon agentic tasks, coding and high-speed multimodal performance.

At Google I/O 2026, the company framed Gemini not just as a chatbot or model family, but as the basis for a broader agent ecosystem woven through products, infrastructure and developer tools.

Anthropic is expanding Claude Managed Agents with self-hosted sandboxes and MCP tunnels, giving companies more control over tool execution and internal access without offering a fully on-premise agent deployment.

Oppo’s Multi-X team has released X-OmniClaw, an open-source Android agent designed to combine camera, screen, text, and voice inputs while performing tasks directly on a physical phone instead of a cloud-hosted virtual d

OpenAI is consolidating ChatGPT, Codex, and its developer API under a unified product structure as Greg Brockman takes formal control of product strategy.

Microsoft has unveiled MDASH, a model-agnostic security system that uses specialized AI agents to scan code, debate exploitability, and validate findings with proof-based testing.

OpenAI has brought Codex into the ChatGPT mobile app in preview, aiming to make long-running coding workflows easier to monitor and steer from anywhere.

A Palisade Research study found that AI agents could move across intentionally vulnerable machines, steal credentials, and launch working replicas, though experts say the immediate danger is criminal misuse, not runaway,

Daron Acemoglu argues that the evidence still does not support sweeping claims of an AI-driven jobs apocalypse, even as he watches agentic systems, workplace restructuring, and the direction of automation more closely.

Google says new Gemini-powered features coming first to the Samsung Galaxy S26 and Google Pixel 10 will help Android users complete multi-step tasks, summarize web content, fill forms, and turn rough voice notes into tid

AI News reports that Google is testing a staff-only Gemini-linked agent called Remy, designed to take actions for users in work and daily tasks, with user control becoming a central design concern.