🤖 China builds AI that builds itself

Good Morning! This is genuinely fascinating, Chinese AI startup MiniMax just dropped M2.7, a proprietary frontier model that literally helped build itself. Plus, I’ll show you how to use Claude Cowork from you're phone.

Plus, in today’s AI newsletter:

• MiniMax has released M2.7

• Perplexity Brings Comet Browser to iOS and iPadOS

• The AI Spending Flip: Anthropic Overtakes OpenAI

• How to use Claude Cowork from you're phone

• 4 new AI tools worth trying

AI MODELS

🤖 MiniMax M2.7: The AI That Builds Itself

MiniMax has released M2.7, a highly efficient reasoning model that marks a massive shift toward recursive self-improvement in AI. Instead of relying purely on human engineers, MiniMax had the model actively participate in its own development cycle.

• M2.7 autonomously handled 30-50% of the team's reinforcement learning research workflow, including experiment monitoring, debugging, and merge requests.

• The model ran 100+ autonomous loops where it analyzed its own failure trajectories, modified its scaffold code, and ran evals, resulting in a 30% performance improvement on internal benchmarks.

• It proves its capability in the wild, too: over three 24-hour trials on MLE Bench Lite (22 ML competitions), M2.7 trained models that earned a 66.6% medal rate, tying Google's Gemini 3.1.

• It delivers intelligence comparable to GLM-5 but at less than one-third the cost ($0.30 per 1M input / $1.20 output tokens).

We are officially crossing the threshold into recursive self-improvement. When AI models can monitor, debug, and optimize their own architectures, the bottleneck for AI advancement shifts from human engineering hours to raw compute power. If a model is already doing 50% of the research workflow today, a fully autonomous AI research lab isn't just sci-fi anymore, it's the immediate future.

AGENTIC AI

📱 Perplexity Brings Comet Browser to iOS and iPadOS

Perplexity’s Comet AI browser is officially live on iOS and iPadOS today, four months after its Android debut. The mobile browser comes packed with a built-in AI assistant, a native ad blocker, and agentic capabilities that currently outpace mobile competitors like Chrome and Edge.

• Comet uses Perplexity’s answer engine by default, allowing users to ask follow-up questions, use voice commands to query web pages, and instantly summarize product reviews.

• The built-in personal assistant can automate complex tasks, research the web, and organize emails using a virtual cloud browser that actively reports back with updates along the way.

• The browser learns your habits over time to help you seamlessly pick up right where you left off.

• Perplexity also dropped a dedicated Prompt Library on its website to help users unlock the full potential of the browser's agentic features.

Bringing true agentic AI to mobile is a massive shift in how we interact with our phones. Instead of just searching for information or passively scrolling pages, users can now delegate complex, multi-step workflows directly from their iPhones. If Comet delivers on its promises, it could seriously challenge Safari's native dominance by turning the mobile browser into an active, autonomous assistant.

AI NEWS

💼 The AI Spending Flip: Anthropic Overtakes OpenAI

The AI race is fundamentally shifting from who has the smartest model to who can monetize the fastest, and Anthropic is pulling ahead with the enterprise customers that actually pay the bills.

• According to Ramp data, Anthropic is now capturing over 73% of all spending among companies buying AI tools for the first time.

• This is a brutal reversal for OpenAI: just 10 weeks ago, the split was 50/50, and as recently as early December, OpenAI was leading 60/40.

• While OpenAI is still on pace to generate more total revenue this year ($25 billion vs. Anthropic's $19 billion), they are actively losing money subsidizing the token costs for their massive consumer base.

• The financial pressure is forcing a pivot: OpenAI is reportedly stepping back from broad consumer bets (like video generators and browsers) to tighten its focus strictly on the enterprise market.

The AI industry is building the plane while flying it, but they are quickly realizing they need more business-class seats. Anthropic's surging momentum proves that massive enterprise adoption isn't won through consumer hype, it's won by proving you can safely and reliably handle corporate workflows.

HOW TO AI

🗂️ How to use Claude Cowork from you're phone

In this tutorial, you will learn how to set up Claude Co-work Dispatch, allowing you to text Claude from your phone and have it perform complex, file-level tasks on your physical computer while you are away from your desk.

🧰 Who is This For

• Researchers looking to automate operations

• Content creators needing an assistant to compile research

• Busy professionals who want their AI to finish compiling spreadsheets while they are hitting the gym for a workout

• Anyone who wants the remote power of an agent like OpenClaw without the complex technical setup, server hosting, or security risks

STEP 1: Prepare Your Desktop Environment

To start, ensure you have the Claude desktop app installed and are on the Pro or Max plan.

Once you’re ready, open the app, navigate to the Co-work tab, and click the new "Dispatch" button on the left-hand menu. The most crucial part of this step is to toggle on "Keep this computer awake", Dispatch cannot operate if your PC goes to sleep or is closed. Finally, grant Claude access to your local files and enable all browser actions so it can navigate pages and fill out forms on your behalf.

STEP 2: Pair Your Mobile Device

With your desktop ready, open the Claude mobile app on your phone. Tap the Dispatch icon and click the "Pair with desktop" button. In less than 60 seconds, your phone is now securely tethered to your physical computer.

You do not need to configure API keys or set up background servers. As long as your desktop remains open and awake at home or the office, Claude will execute any commands you text it from your phone.

STEP 3: Connect Your Apps and MCP Servers

To make your remote agent truly powerful, it needs access to your daily software. On your desktop, go to your Connectors and enable native integrations like Gmail, Slack, and Google Calendar. If you need to connect to apps outside of Claude's default 38 integrations, use the Zapier MCP (Model Context Protocol). By building a Zapier MCP server and pasting its URL into Claude’s connector settings, you instantly give your Dispatch agent the ability to execute actions across 8,000+ different applications.

STEP 4: Trigger Automations on the Go

Now you can manage your digital life entirely from your phone. Text your mobile app a prompt like, "Find the receipts in my local Expenses folder and give me a breakdown," and Claude will search your computer's hard drive and text you the itemized list. Or, ask it to run a "Morning Brief," where it scours your unread emails and calendar events to text you your top priorities for the day. Your desktop does all the heavy lifting, and your phone gets the polished results.

Facebook launches Creator Fast Track, offering big Instagram, TikTok and YouTube creators guaranteed monthly pay and boosted reach to post on Facebook.

Meta confirms a critical security incident after an internal rogue AI agent's actions led to the exposure of sensitive data to employees without authorization.

Anthropic releases a survey of 80,508 Claude users' views, hopes, and fears about AI, calling it “the largest and most multilingual qualitative study”.

🤝 MiniMax M2.7: A proprietary frontier model that literally helped build itself

🤏 GPT-5.4 Mini & Nano: fast, cheap AI models made for coding agents

⚙️ OpenAI Codex: OpenAI’s coding assistant, now with automations and customizable themes

🎥 Replit Animation: Turn text prompts into animated videos

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas