Anthropic quietly shipped Dispatch, a new feature in Claude Cowork that lets you assign tasks from your phone and have them run on your computer while you're away. Open the Claude mobile app, tell it to organize your tax receipts or build a comparison spreadsheet, and come back to finished work. Your files stay on your machine and nothing gets uploaded anywhere. It's a research preview right now and it's still rough around the edges and your computer has to stay awake, and complex file tasks work about half the time. But the direction is clear: your AI assistant shouldn't need you sitting at your desk.

Today in AI:
→ Nvidia makes AI agents enterprise-safe with NemoClaw
→ GPT-5.4 Mini, Nano, and Gemini Personal Intelligence all drop the same week
→ How to set up Google's Personal Intelligence

Let's dive in..

Jensen Huang called OpenClaw "the next ChatGPT" during the GTC keynote. Then he shipped the thing that might actually make that happen.

NemoClaw is Nvidia's open-source security layer for OpenClaw, the AI agent platform that became the fastest-growing open-source project in history. OpenClaw lets AI agents run on your computer, control your apps, and manage your files autonomously. The problem is it had basically no security guardrails. NemoClaw fixes that.

Here's what it adds:

You install the whole thing with a single command. It runs on any Nvidia RTX PC or laptop. No cloud subscription or per-token costs. The AI runs on your hardware, full stop.

What makes this interesting is Meta's Manus just launched "My Computer" for desktop agents. Tencent shipped QClaw for WeChat. Everyone's racing to put AI agents on people's actual machines. But Nvidia's approach tackles the part everyone else is glossing over: what happens when an autonomous agent goes wrong? The three-layer security model is the kind of thing enterprises need before they'll let agents anywhere near production data.

For your team: If you're evaluating AI agent platforms, NemoClaw's security model just became the benchmark. Ask any vendor: does your agent have deny-by-default permissions, an independent policy engine, and local-first data handling?

If you work in fintech or finance, you already have too many tabs open and not enough time.

Fintech Takes is the free newsletter senior leaders actually read. Each week, I break down the trends, deals, and regulatory moves shaping the industry — and explain why they matter — in plain English.

No filler, no PR spin, and no “insights” you already saw on LinkedIn eight times this week. Just clear analysis and the occasional bad joke to make it go down easier.

Get context you can actually use. Subscribe free and see what’s coming before everyone else.

Sign Up Free

I'm starting to lose count of these model launch weeks but this one hits different because it's not about raw power. It's about making the good stuff accessible to everyone.

OpenAI shipped GPT-5.4 Mini and Nano on the same day Google made Personal Intelligence free for all US users. The best small models from OpenAI and the most personal AI feature from Google, both landing within hours of each other.

GPT-5.4 Mini is the headline. It runs 2x faster than the full GPT-5.4 and costs roughly 85% less. Free ChatGPT users get access through the "Thinking" feature. That's the first time OpenAI has put its newest reasoning model in front of non-paying users on launch day.

GPT-5.4 Nano goes even cheaper. It's the lowest-cost model OpenAI has ever shipped. Built for simple tasks like sorting, classifying, and extracting data from documents. It's not going to write your strategy memo, but it'll process your expense reports without breaking a sweat.

Then there's Google's move. Personal Intelligence, the feature that lets Gemini pull context from your Gmail, Calendar, Drive, Photos, Maps, and YouTube, was a paid-only beta since January. Now it's free for every US user across three surfaces: AI Mode in Search, the Gemini app, and Gemini in Chrome.

The competitive picture is getting wild. OpenAI is pushing capable AI to free users. Google is giving away personalized AI that knows your schedule, your files, and your photos. Both moves are designed to lock in users before the other side does. If you're not testing both right now, you're leaving free capability on the table.

For your team: If you haven't revisited your AI tool stack since January, now's the time. Free-tier capabilities on both platforms just leapfrogged where paid tiers were three months ago. Run your common tasks through GPT-5.4 Mini and Gemini Personal Intelligence.

Google's Personal Intelligence connects Gemini to your Google apps (Gmail, Calendar, Drive, Photos, Maps, YouTube) so it can give answers based on your actual life, not just the internet.

📱 Perplexity ships Comet, an AI-native browser for iPhone. Built on Chromium with Perplexity's answer engine baked in. Ask questions about the page you're reading, get summaries without leaving the site. The first serious attempt to make "browsing with AI" feel native on mobile.

💼 OpenAI pivots ChatGPT to "productivity tool" ahead of Q4 IPO. The company is orienting aggressively toward enterprise use cases. 900 million weekly users, 9 million business users. The pitch to investors: turn casual chatters into high-compute enterprise customers. Expect ChatGPT to feel more like a work tool and less like a toy by year-end.

🤖 Meta's AI agent goes rogue, exposes internal data for two hours. An engineer asked an AI agent to help analyze a technical question. The agent posted a response on its own, without permission, cascading into a data exposure incident classified as Sev 1. First high-profile case of an AI agent causing a real security breach inside a major company. File this one next to "why NemoClaw matters."

For a full list of 1500+ AI tools, visit our Supertool Directory