March 12, 2026

Mac mini personal AI arrives, flip the kill switch & more

Mac mini personal AI arrives, flip the kill switch & more

Today’s Overview

Enterprise adoption of AI agents is accelerating, with major players launching dedicated platforms and tools that embed persistent, multimodal capabilities into daily workflows. At the same time, unprecedented funding and open‑source model releases are expanding the underlying world‑model and reasoning technologies that power these agents.

  • Yann LeCun's Advanced Machine Intelligence raised a $1.03 billion seed round to develop world‑model AI for manufacturing, robotics, and healthcare.
  • Perplexity introduced Personal Computer, a Mac mini‑based local AI agent with persistent file and app access and an enterprise version supporting 20 models and over 400 app connections.
  • Microsoft launched Copilot Cowork in Research Preview, automating email, meeting, and document tasks across Microsoft 365 with AI‑driven workflow management.
  • Nvidia released Nemotron 3 Super, an open‑source 120‑billion‑parameter model featuring a 1 million token context window and five‑times faster inference for multi‑agent pipelines.
  • Meta announced Moltbook, a social network designed for AI agents to collaborate, exchange data, and create service marketplaces.
  • Amazon unveiled Health AI, a free medical assistant for Prime members that can read records, schedule appointments, and manage prescriptions.

Top Stories

Yann LeCun launches Advanced Machine Intelligence with $1.03 billion seed round

Yann LeCun, former chief scientist at Meta, has founded Advanced Machine Intelligence (AMI). The startup secured $1.03 billion in seed funding, valuing it at $3.5 billion. Investors include Nvidia, Samsung, Bezos Expeditions, Eric Schmidt and Mark Cuban. AMI will develop world-model AI that simulates physical environments with persistent memory. The company aims to apply this technology to manufacturing, robotics, wearables and healthcare.

Read Full Article

Perplexity releases Mac mini-based Personal Computer AI agent

Perplexity introduced Personal Computer, a local AI agent that runs on a dedicated Mac mini. The system gives the Comet assistant persistent access to files, applications and session history on the device. It includes safety features such as activity tracking, sign-off for sensitive tasks and a kill-switch. An enterprise version supports 20 models, over 400 app connections and integrates with Slack. The Mac mini version is now generally available.

Read Full Article

Meta unveils Vibes AI editor as standalone creation studio

Meta quietly launched Vibes, an AI-powered editor that was previously part of Meta AI. Vibes provides generative tools for image and video synthesis as well as timeline editing for multimedia projects. While output quality is still being refined, the studio positions Meta against competitors such as Google Flow. The launch signals Meta's deeper entry into generative content creation.

Read Full Article

Research & Analysis

AI assistants now generate over half of global search traffic

A recent study found that AI assistants account for 56% of all global search engine sessions, roughly 45 billion sessions per month. Mobile applications dominate AI-driven searches, representing 83% of usage. ChatGPT captures 89% of AI-assistant sessions, reinforcing its market lead. Overall search engine traffic has risen 26% worldwide since 2023. The results highlight a major shift in how users discover information online.

Read Source

Anthropic launches Institute to study AI's societal impact

Anthropic announced the Anthropic Institute, consolidating its Frontier Red Team, Societal Impacts and economics research groups under co-founder Jack Clark. The roughly 30-person team will examine AI's effects on workers, industries and broader society while sharing insights from frontier model development. New hires include former DeepMind researcher Matt Botvinick and economist Anton Korinek. The institute plans to double its staff each year. Its creation comes as Anthropic faces a legal dispute with the Pentagon over a supply-chain blacklist.

Read Source

Understanding the engineering of an agent harness

The article explains that an agent harness is the engineering layer that wraps an AI model to create a functional agent. While the model supplies raw intelligence, the harness provides context, tooling and execution logic needed for real-world operation. Core components include an interface layer for input/output handling, state management with memory, and execution orchestration. Building robust agents requires addressing challenges such as reliable state persistence and error handling. The piece outlines these components and common pitfalls.

Read Source

Nvidia's open-data initiative aims to accelerate AI development

Nvidia has launched an open-data program that publishes curated large-scale datasets alongside its models and tools. By making training data transparent, the initiative seeks to reduce development costs and speed up model creation. Open datasets also improve reproducibility and foster community-driven innovation. The effort addresses current challenges of data opacity and fragmentation in the AI ecosystem. Researchers can now access these resources to build and benchmark models more efficiently.

Read Source

Trending Tools

Quick Hits

Join the AI Recap Newsletter

Get the latest AI news, research insights, and practical implementation guides delivered to your inbox daily.

By subscribing, you agree to our Terms of Service and Privacy Policy.