What's Actually Happening
Your Mac just got a second pair of hands. Google is rolling out computer use in the Gemini app for Mac, so Gemini can click, type and fill in forms across your apps, files and websites in a background window while you keep working.
It lands the same day a Bloomberg report says Apple is hiring to build its own Mac agent, and is adding warnings that discourage people from handing outside agents full control. The fight over who gets to drive your desktop has officially started.
Also tonight: OpenAI ships GPT-Live-1.1 for developers, Ideogram 4.6 takes the top spot for multi-step image editing, an Anthropic mathematician claims a Riemann Hypothesis proof made with Claude, and four tools worth trying.
AI AGENTS
🖥️ Gemini Can Now Use Your Mac For You
Google announced on November 19 that the Gemini app for Mac can now operate your computer. Add @Computer use to a prompt and Gemini works through native Mac apps, local files and the web, navigating menus, clicking, typing and filling in forms to finish multistep jobs. It runs in an isolated background window with a live preview inside the chat, so your own screen and cursor stay free.
Google's examples are the tedious kind: batch color-grading photos in design software, pulling weights and addresses out of local PDFs to fill in shipping quotes, building a Google Slides deck from live dashboards, and turning a folder of clothing photos into priced resale listings. You choose which apps Gemini can touch, can pause or stop it at any time, and it stops for logins and payments and asks before sensitive actions. It's rolling out in English in the U.S. to Google AI Plus, Pro and Ultra subscribers (18+, personal accounts) on Apple Silicon Macs running macOS Sequoia 15 or later.
Our take: Google is late here. OpenAI, Anthropic and Meta already ship desktop agents, as TechCrunch points out. But the background window is the right design choice: an agent that borrows your cursor is a demo, while one that works beside you is a coworker. Start with one boring weekly chore you can check by eye, and keep it away from anything that sends money or messages until you've watched it work a few times.
Will you let Gemini run tasks on your Mac this week?
Go deeper: Download Gemini for Mac · How to turn it on (9to5Google)
🍎 Why It Matters: The Fight For Your Desktop
The same day Google shipped this, Bloomberg spotted an Apple job listing for a senior engineering manager to build an AI agent that can use a computer: browsing the web, working across apps, running commands and writing and debugging code. Bloomberg also reports macOS 27.3 will warn users against giving outside agents full access to their Macs, after Apple told developers in October it would require explicit user action for that level of control. Apple has reportedly eyed a launch as soon as the end of next year, though the work is early, according to TechCrunch.
Put those together and the shape is clear. Every big lab now wants its agent to be the one that drives your computer, and the company that owns the operating system decides how hard it is to say yes. Expect more permission prompts for Gemini, Claude and ChatGPT agents on the Mac before Apple's own arrives. The privacy stakes are not hypothetical either: Reuters reported this week that Meta employees saw its Muse agent surface other people's bank balances and orders, which Meta says involved internal test versions. Grant agents the fewest apps they need and check what they can reach.
A Word From Today's Sponsor
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
Let’s Talk MODELS! 🤖
OpenAI released GPT-Live-1.1, an updated full-duplex voice model for apps that talk to people in real time. Full-duplex means it can listen while it speaks, so callers can interrupt naturally. OpenAI says it transcribes speech more accurately than GPT-Live-1, keeps a more consistent accent and voice across languages, and lets you describe in plain language when it should hand a task to your backend for reasoning or tool calls.
Pricing is per minute of session time: $0.05 while a session is active and $0.02 on standby, when a WebRTC session is paused. Backend model and tool use are billed separately. It supports streaming and function calling and runs over the Live and Realtime endpoints.
Our take: per-minute pricing is the detail builders should care about. It makes a voice agent's cost as predictable as a phone plan, and the cheap standby rate means you can keep a line open for hold music or long silences without burning money. Run a few real calls with accents your customers actually have before you switch.
Try it: Model docs · Live sessions guide · API pricing
The Image Model That Edits Without Breaking Things
Ideogram released version 4.6 of its image model on November 19. It generates natively at up to 4K, not upscaled, and its headline feature is multi-turn editing: you keep refining the same image over several rounds, and Precise Edit accepts an optional mask so changes stay where you point them. Generate also takes up to five reference images.
Ideogram says 4.6 ranks first on Rapidata's Multi-step Image Edit leaderboard with a score of 1,892, just ahead of its own 4.5 at 1,866. In the API, images start at $0.008 at the lowest quality (up to 2K) and run to $0.20 for a high-quality 4K generation and $0.60 for a high-quality 4K edit.
Our take: the hard part of AI images was never the first draft, it's the fifth revision where the model quietly redraws the face you liked. A leaderboard built on multi-step edits measures exactly that, and the lead over 4.5 is narrow, so test it on your own revisions. The sub-cent tier makes it cheap to try.
Try it: Ideogram app · 4.6 API endpoints & pricing
Top 5 In AI Research 🔬
The other stories worth your attention:
An Anthropic mathematician claims a Riemann Hypothesis proof made with Claude, posted with a machine-checkable Lean archive for others to verify. Unconfirmed: independent mathematicians have not checked it yet. See what still has to be verified. (Scientific American)
Anthropic closed its IPO bigger than the headline number after underwriters bought every extra share they were allowed to, at the same $83 price. See the final tally. (Anthropic)
Genie 4 is now in the Gemini API, so developers can stream playable, prompt-editable 720p worlds with native audio into their own apps. See how to start building. (Google)
OpenAI lets Codex agents hunt down ChatGPT's slowdowns on their own, running investigations around the clock, and published how often their diagnoses hold up. See the hit rate. (OpenAI)
Gemini agents can now load tools only when they need them, a new preview in the Gemini API aimed at agents with dozens or hundreds of tools. See the trade-off Google says to measure first. (Google)
🛠️ Tools That Are Hot Right Now!
🎙️ ChatGPT Live Dictation - talk to ChatGPT and watch your words appear as you speak, then edit them before you send. Free on every ChatGPT plan, on web, desktop, iOS and Android; see which old limit OpenAI just dropped.
🔐 Devin Security Threats - scan your repos for newly disclosed vulnerabilities and have Devin open fix pull requests, with auto-scans when new threats are published. Part of Devin (paid; Cognition hasn't listed a separate price); see the two threats it launched with.
🔌 Antigravity SSH - run Google Antigravity's coding agents on a remote Linux box, so they keep working after your laptop sleeps. Built into the app (v2.33.0+) at no extra listed cost; see what your server needs.
🤖 NVIDIA MotionBricks - animate game characters and robots in real time by placing target poses, with demos and scene editors included. Free and open source (needs an NVIDIA GPU); see how fast it generates motion.
What's The Recap?
Google gave Gemini for Mac the ability to use your computer, working through apps, files and websites in a background window for U.S. Google AI Plus, Pro and Ultra subscribers, while Bloomberg reports Apple is building its own Mac agent and adding warnings for outside ones. OpenAI shipped GPT-Live-1.1, a listen-while-it-talks voice model at 5 cents a minute, and Ideogram 4.6 took first place on a multi-step image-editing leaderboard. An Anthropic mathematician claimed a Riemann Hypothesis proof made with Claude, still unverified, and Anthropic's IPO closed larger than first reported. The thread: AI is moving from answering questions to doing the work on your own machine, and the next fight is over who you let in.
What Do You Rate Today's Newsletter?
Stay building. 🤖
We put Gemini 4 Neon through a full test against Opus 5.5 and GPT-6.2.
📬 Read by 14K AI readers. Know someone who'd like it? Send them here to subscribe.



