Gemini Can Now Use Your Mac for You: Google Adds Computer Use to the Gemini App
Describe a job, and Gemini now does the clicking itself — navigating your Mac apps and websites in a background window while your own screen and cursor stay free.
103 prompt sets · 539 copy-paste prompts
21 Academy lessons live, free
3 new posts a day
New AI news articles as they break
The news desk
Source-linked briefings about AI at work, in creative production, and at the research frontier. Each includes a practical next step and an explicit limitation.
Publisher headlines
These are headlines supplied by the original publishers, not independently verified A.I. Intel Desk reporting. Open the source before relying on a claim.
Checking approved sources. If the live feed is unavailable, the reviewed briefings below remain available.
The A.I. Intel Desk briefings below are published first on the A.I. Intel Desk Substack and mirrored here as they appear — this page shows the last five days of briefings. Publication dates belong to the original stories. Company announcements are labeled and have not been independently validated as performance results.
Describe a job, and Gemini now does the clicking itself — navigating your Mac apps and websites in a background window while your own screen and cursor stay free.
Google just put a frontier-level model in the Pro tier: Gemini 4 Neon scores 61 on the Artificial Analysis Intelligence Index — #3 overall, just behind larger, slower models — and at its default setting it costs 32% less per task than Carbon, Google's own figures show.
Google's Genie 4 builds explorable worlds from a prompt — and starting today, paying subscribers can step inside one, rewrite the world while they walk through it, and hear it respond with sound and dialogue generated live.
Perplexity's new Multiagent level puts a Claude Fable 5.2 lead in charge of agents running three rival labs' models — working in parallel, sharing evidence and flagging each other's mistakes before one answer comes back.
Microsoft's new flagship coding models are live in GitHub Copilot today — and Microsoft says the big one delivers 98% of GPT-6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5's performance at just 5%.
Google just made Gemini's task mode free for hundreds of millions of people — and in the U.S. it can now call businesses for you, pay with tokenized cards, take instructions by text and turn forwarded emails into calendar events.
OpenAI's new frontier model tops its own professional-work test while costing less than a third as much per task — and its cheaper sibling nearly matches it on real code at about a fifth of the cost.
Anthropic's own help page now lists the four holdout models as watermarked — the second wave that was only a customer-email promise on November 4 is confirmed on a company page today.
OpenAI's agents can now read screens, click and type across the health record and insurer portals your care team already uses — drafting the order and the prior authorization for a human to review, instead of staff copying the same details between systems by hand.
Google DeepMind announced AlphaGenome Atlas, a resource containing predictions about the effects of genetic variants. The announcement places AI inside a research workflow rather than a general-purpose chat experience.
OpenAI published company examples of agents used in onboarding, account work, and developer integrations. The cases emphasize repeatable steps, relevant context, and ways to review the work.
Anthropic announced an early research preview of the Model Hardware Standard, intended to help agents interface with scientific and manufacturing equipment.
Go to the primary source
Four places to begin a source check. These links open the original publishers; they are not a live feed or a claim that every article has been reviewed.
Product and research announcements from the provider. Separate its reported results from independent evidence.
Open the original sourceGoogle’s AI news and explainers. Check the product, plan, region and date that apply to the article.
Open the original sourceResearch, community and organization-authored articles. Check who wrote each piece rather than treating the host as its author.
Open the original sourceResearch explanations from Microsoft. A research announcement is not proof that a product is ready for your workflow.
Open the original source