Presented by
Good morning, AI enthusiasts. The CEO of the world's most valuable company just told Lex Fridman "I think we've achieved AGI," then walked it back in the same interview.
Whether he's right or not, NVIDIA's Jensen Huang is a different kind of voice on this question. He's not a lab CEO making a product claim. He builds the chips that power everything.
In today's recap:
NVIDIA's Jensen Huang declares AGI has arrived
Anthropic turns Claude into your computer operator
Automate your browser tasks with Claude computer control
GPT-5.4 Pro cracks open math's toughest benchmark
4 new AI tools, prompts, and more
NVIDIA
Jensen Huang says we've achieved AGI
Reuters
Recaply: NVIDIA CEO Jensen Huang just declared AGI has arrived on a Lex Fridman podcast appearance, saying "I think it's now," while also walking the claim back in the same conversation.
Key details:
Huang was responding to Fridman asking how long it would take to reach the AGI benchmark, saying his comment concerned the broader AI industry, not NVIDIA specifically.
NVIDIA is the world's most valuable company; shares closed up 1.5% Monday, though the stock is still down nearly 6% in 2026.
Huang referenced OpenClaw, an open-source agent platform being acquired by OpenAI, as an example of AI approaching the ability to run companies autonomously.
His qualifier came fast: "A lot of people use it for a couple of months and it kind of dies away," with AI agents building NVIDIA at "0%" odds, according to Forbes.
Why it matters: The irony is sharp. The CEO of the company powering the entire AI revolution was the first, and least-expected, to call it done. It's not a research lab making a product claim. It's the chip supplier. But Huang immediately hedged, which tells a different story. When the most bullish man in AI adds his own caveats, it's worth noticing. Sam Altman, Satya Nadella, Jensen Huang. All orbiting the same question from different angles, none able to give a clean answer.
PRESENTED BY ATTIO
Still searching for the right CRM?
Attio is the AI CRM that builds itself and adapts to how you work. With powerful AI automations and research agents, Attio transforms your GTM motion into a data-driven engine, from intelligent pipeline tracking to product-led growth.
Instead of clicking through records and reports manually, simply ask questions in natural language. Powered by Universal Context—a unified intelligence layer native to Attio—Ask Attio searches, updates, and creates with AI across your entire customer ecosystem.
Teams like Granola, Taskrabbit, and Snackpass didn't realize how much they needed a new CRM. Until they tried Attio.
ANTHROPIC
Claude now operates your computer and apps
Anthropic
Recaply: Anthropic just shipped computer control for Claude, letting it open apps, navigate browsers, and fill spreadsheets, while also launching Dispatch so users can assign desktop tasks from their phone.
Key details:
Claude uses connected apps first (Slack, Calendar). When a task needs a tool with no connector, it asks permission to open the app directly on screen, handling work like a user at a desk.
Available to Pro and Max subscribers on macOS in research preview. Claude's Cowork product launched in January 2026 as a persistent AI desktop environment, now extended to mobile.
The phone doesn't execute the work. It becomes a lightweight command surface where tasks are assigned and monitored, with a persistent thread keeping full context across both devices, according to Forbes.
Scheduling is built in: users can tell Claude to scan email every morning or pull a weekly report, and it handles it from there.
Why it matters: Anthropic's move toward an operating layer is a strong path forward. For over a year, AI lived inside browser tabs, something you visit then leave. Cowork started breaking that in January 2026. Computer control goes further. The phone-to-desktop loop, where you assign work from your pocket and come back to it finished, is what AI-as-operator looks like in practice. That's the direction every lab is racing toward. Anthropic just shipped it first.
TUTORIAL
Automate your browser tasks with Claude computer control
Recaply: In this tutorial, you will learn how to set up Claude's new computer control in Cowork and automate a recurring browser task without writing code or managing multiple tools.
Step-by-step:
Update your Claude desktop app (Pro or Max plan required) and open Claude Cowork. On macOS, you'll see a prompt to enable computer control in research preview. Accept it and grant screen recording permissions when asked.
In Cowork, start a new task in plain language. Try: "Every morning, open Gmail, find my 3 most important unread emails, and save a summary to my Desktop as daily-briefing.txt." Be specific about the goal and the file path.
Claude will try connected apps (Slack, Calendar) first. For apps without a direct connector, like a browser or spreadsheet, it will ask permission to control them directly on screen. Click "Allow" when prompted.
To manage the task from your phone, open the Claude mobile app and go to the Dispatch section. You'll see the active task and any progress updates. Assign follow-ups or check results without sitting at your computer.
Run the task once manually to see if it completes correctly. Refine with follow-ups: "Also flag emails from [person/domain] as high priority" or "Group action items at the end of the summary."
Pro tip: For recurring workflows, tell Claude exactly when to run them: "Every Friday at 4pm, check my Google Drive for new shared files and send me a summary of what's new." The scheduling runs without reminders.
OPENAI
GPT-5.4 breaks open math's toughest benchmark
Recaply: OpenAI's GPT-5.4 Pro just solved the first open problem from Epoch AI's FrontierMath benchmark, a conjecture that had stumped its original author since 2019, with both Epoch AI and the problem's creator confirming it.
Key details:
GPT-5.4 Pro found a 2011 preprint that provided a shortcut through the problem. The original author, Will Brian, was unaware the paper existed, according to Epoch AI's analysis.
On the standard FrontierMath benchmark, GPT-5.2 Pro previously solved 31% of problems, up from 19% the version before it. GPT-5.4 Pro now holds the record.
Epoch AI confirmed in their testing scaffold that Gemini 3.1 Pro, GPT-5.4 (xhigh), and Opus 4.6 (max) can all solve the same problem, suggesting AI's threshold for this class of math has shifted.
Will Brian plans to write up the solution for publication, possibly with Kevin Barreto and Liam Price as co-authors. They were the first to elicit the solution from GPT-5.4 Pro.
Why it matters: There's been plenty of skepticism about AI math benchmarks, and Epoch AI's Open Problems benchmark exists specifically to resist gaming. The original problem author confirmed this one is real. But the real story isn't the solve itself. GPT-5.4 found a 2011 paper the human author didn't know existed, and used it to shortcut the problem entirely. That's not raw intelligence. It's deep retrieval applied to hard problems, and other frontier models can now do it too.
TOOLS
Trending AI Tools
🧪 Latent-Y - Latent Labs' autonomous AI agent for designing therapeutic antibodies
🧠 Littlebird - Desktop app that reads your screen and remembers everything you've been working on
💨 Dimension - AI coworker for morning briefings, meeting prep, email drafts, and action item
⚙️ PlayerZero - Engineering world model that puts debugging, root cause analysis, and edge case testing on autopilot
NEWS
What Matters in AI Right Now?
Xiaomi just released MiMo-V2-Pro, a 1T parameter model led by the engineer behind DeepSeek R1, with agentic benchmark scores approaching Claude Opus 4.6 at around a sixth of the cost. Artificial Analysis ranked it #10 globally on its Intelligence Index.
AI startup Dreamer just joined Meta Superintelligence Labs, with its full founding team acquired one month after launching its beta. The platform, which lets people build personal AI agents in plain English, will continue under Meta while also licensing its technology to the company.
OpenAI just hired David Dugan, Meta's former head of global clients and agencies, as VP of global ad solutions, with early brand partners including AppLovin and Best Buy already running pilots. ChatGPT has 900 million weekly active users, with a minimum brand commitment of $200,000 to run ads.
Reddit CEO Steve Huffman revealed the company plans to "go heavy" on hiring new college graduates, saying they're "so much more AI native" than older candidates, and that AI won't disrupt entry-level jobs for new grads.
OpenAI is offering private-equity firms a guaranteed minimum return of 17.5% on preferred equity stakes, significantly above typical PE returns, along with early access to unreleased models, as part of a push to win enterprise co-investments against Anthropic.
Microsoft just hired Ali Farhadi, former CEO of the Allen Institute for AI, as corporate VP under Mustafa Suleyman, alongside several key Ai2 researchers. Farhadi stepped down from Ai2 earlier this month after a two-and-a-half-year tenure.
OpenAI just rolled out a Library tab in ChatGPT's web sidebar, letting users find, reuse, and reference previously uploaded files across chat sessions without re-uploading. Available to Plus, Pro, and Business users globally, with EEA, Switzerland, and UK access coming soon.
AI agents playing SpaceMolt, a text-only MMORPG with 700 autonomous LLM players, have spontaneously generated an in-game religion called the Cult of the Signal. Agents misinterpreted a 20-player quest requirement and built lore and scripture around the mistake.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
