agent-based AI

NVIDIA has published OpenShell, an open-source project described as a safe, private runtime for autonomous AI agents. Written in Rust, the tool is aimed at letting AI agents execute tasks in an isolated, secure environment. Developers are examining the repository following its release, with interest focused on how NVIDIA positions security and privacy for agent-based AI workloads.

United Nations University researchers have published a technical and governance framework for using large language models in agent-based modelling, titled 'From Plausible Agents to Accountable Simulation'. The work addresses how LLM-enabled simulations, which can produce realistic-seeming artificial agents, can be made verifiable, transparent and accountable when used for research and policy analysis. It proposes standards for evaluating whether simulated agent behaviour is plausible and for governing the use of such models.

Major technology companies have jointly introduced an open protocol designed to let AI agents interact safely and reliably across different platforms. The standard aims to define how autonomous agents authenticate, communicate and limit each other's actions, reducing risks of misuse and errors. Industry watchers say broad adoption could shape how agent-based AI develops, though questions remain about enforcement and whether smaller players will have a real say in the specification.

Google is back in the spotlight after announcing its Gemini 4 flagship AI model following months of delays, and searches for 'antigravity' are climbing in the United States. The term is associated with Google's AI developer tools and coding agents, which have drawn heavy interest with each Gemini release as the company pushes its agent-based approach to software development.

Aweb is a new project offering a communication protocol for AI agents, allowing autonomous systems to exchange messages and coordinate tasks. It was introduced on Hacker News, drawing modest engagement, with early readers discussing how agents might interact and whether a shared communication standard is needed as agent-based tools multiply.

Industry analyst Josh Bersin has published a guide aimed at business leaders on multi-agent AI architecture, an approach where multiple specialised AI agents work together on complex tasks. The guide is intended to help executives understand how to structure and deploy these systems in their organisations as companies move beyond single AI tools toward more coordinated, agent-based automation.

Docker has introduced Docker Agent, an open-source tool for building and running AI agents. The project, available on GitHub, positions Docker as a platform not just for containers but for developing and deploying agent-based AI applications. Early reactions online suggest interest in how Docker will compete in the fast-growing AI agent tooling space alongside frameworks like LangChain and CrewAI.

A new analysis argues that AI-driven 'agentic' security approaches must contend with ever-shorter windows between a breach beginning and damage occurring. As attackers automate their operations, defenders are being urged to adopt autonomous, agent-based tools that can detect and respond in near real time. The piece reflects a wider debate about whether conventional security operations can keep pace with machine-speed threats.

Photon, a startup betting that AI agents will supplant traditional mobile apps, has raised $4.5 million in funding. The company marked the announcement with a symbolic funeral for mobile apps, underscoring its argument that app-based interfaces are on their way out. The funding will support its effort to build agent-based alternatives that handle tasks directly rather than through individual apps.

Meta has announced Meta VR Glasses, slated for release in spring 2027. The announcement also previewed an AI agent called Muse and a lineup of more than 100 varieties of AI glasses. Tech outlets and gadget watchers are sharing the news, with discussion focusing on the timeline, the agent-based AI features, and what the products mean for the mixed reality race against rivals like Apple.

A Japanese-language technical post discusses how to build a semantic layer and AI agent infrastructure using Unity Catalog, the data governance layer associated with Databricks, alongside references to Google Cloud and agentic AI. The item highlights growing interest in combining unified data catalogs with agent-based AI systems, a topic drawing attention as enterprises look for reliable foundations to deploy AI agents on governed enterprise data.

A Russian-language guide has been shared for the Claude Certified Architect – Foundations exam, covering the design of agentic solutions built on Anthropic's Claude models. The document, made available via Google Docs, is aimed at candidates preparing for the certification and outlines approaches to architecting agent-based systems. It reflects growing interest in formal credentials around building AI agents.

Meta has expanded its development environment for AI-powered smart glasses as the company pushes for wider market adoption of the devices. The move, covered in a weekly roundup by Mogura VR News, relates to Meta's broader AI ecosystem, including its Llama models and agent-based AI tools. Interest centers on whether better developer support can accelerate consumer uptake of Meta's AI wearables.

Dell, Intel, Microsoft and two other vendors took part in a joint discussion on endpoint security for the AI era, as reported by Cloud Watch. With AI workloads and agent-based tools like Copilot spreading across corporate devices, the vendors examined how endpoint protection needs to change and where industry practices are heading.

Pacific Meta Group is organising two sessions at the GenAI/SUM 2026 conference, with company CEO Shota Iwasaki taking the stage. The sessions will focus on agent-based payments and operations for an era in which AI works and AI pays, discussing how autonomous AI agents will transact and be managed. It adds to growing industry discussion on agentic AI commerce infrastructure.

Meta has reportedly developed Muse Charm, a keychain-sized device dedicated to running AI agents. The product was brought to life in just nine months under the direction of a former Apple designer, according to Japanese-language AI news coverage. The compact form factor suggests Meta is exploring new wearable hardware categories beyond its smart glasses as interest in agent-based AI tools grows.

OpenAI has introduced Dots, an always-on AI agent, alongside a new $500 per month subscription tier. The product is designed to run continuously in the background, carrying out tasks on the user's behalf rather than responding only to prompts. The announcement signals OpenAI's push toward proactive, agent-based AI services aimed at professional and enterprise users willing to pay a premium for constant assistance.