Presented by
Good morning, AI enthusiasts. Cursor launched what it called "frontier-level coding intelligence" last week. A day later, someone found Kimi inside.
Kimi is an open-source model from Moonshot AI, a Chinese company backed by Alibaba. Cursor, a $30B US startup, built its flagship coding model on top of it and said nothing about that in the announcement. When an X user spotted the model ID in the code and called it out, Cursor's co-founder admitted it was "a miss." The question now isn't whether it was legal. It was. The question is why a company worth $30B didn't just say so from the start.
In today's recap:
Cursor's $30B coding model built on Chinese AI
Zuckerberg's AI agent, built to bypass his own team
Set up Claude Cowork Projects for focused work
Pentagon locks Palantir in as US military's AI core
4 new AI tools, prompts, and more
CURSOR
Cursor's new model built on Chinese AI
Cursor
Recaply: Cursor just admitted that Composer 2, its new flagship coding model, was built on top of Kimi 2.5, an open-source model from Moonshot AI, a Chinese company backed by Alibaba, after an X user spotted the model ID in the code.
Key details:
Cursor applied its own reinforcement learning on top of Kimi 2.5 as the base, with only about 1/4 of the compute in the final model coming from Kimi, according to Cursor's VP of developer education Lee Robinson.
Cursor raised $2.3B last fall at a $29.3B valuation and reportedly exceeds $2B in annualized revenue, making it one of the most-used AI coding tools on the market.
Cursor co-founder Aman Sanger admitted: "It was a miss to not mention the Kimi base in our blog from the start." Kimi confirmed the arrangement was an authorized commercial partnership via Fireworks AI.
Composer 2 launched last week; Sanger said the team will disclose base model origins in future model release announcements.
Why it matters: Cursor didn't lie. But it didn't say anything either, which matters just as much in a moment when the AI field constantly frames itself as a US vs. China competition. Building on an open-source Chinese model is legal and properly licensed. But it highlights a real gap between how AI tools get marketed and what's actually powering them. Readers who pay for premium AI products deserve to know what's under the hood. The Kimi reveal won't hurt Cursor's business. But it will make people ask harder questions about the next model too.
PRESENTED BY MORNING BREW
When it all clicks.
Why does business news feel like it’s written for people who already get it?
Morning Brew changes that.
It’s a free newsletter that breaks down what’s going on in business, finance, and tech — clearly, quickly, and with enough personality to keep things interesting. The result? You don’t just skim headlines. You actually understand what’s going on.
Try it yourself and join over 4 million professionals reading daily.
META
Zuckerberg builds AI agent to run his job
Bloomberg News
Recaply: Meta CEO Mark Zuckerberg just started building a personal AI agent to retrieve information faster, replacing the layers of executives he'd normally go through to get answers.
Key details:
The agent is in development. It helps Zuckerberg get answers he'd normally need to ask multiple teams to find. That means fewer people in the information chain.
Meta has 78,000 employees and has been cutting organizational layers. In January, Zuckerberg told investors the company is "elevating individual contributors and flattening teams."
A person familiar with the project said Zuckerberg has been spending more time coding too. It signals he's going deeper on AI tools at the executive level, not just delegating the work.
No public launch date has been set. Meta sees AI adoption as critical to staying competitive against leaner, AI-native startups with much smaller headcounts.
Why it matters: This isn't just about Zuckerberg getting faster answers. It's about where agentic AI is actually heading. When the CEO of a 78,000-person company replaces human layers with AI, these tools have clearly moved beyond software teams. They're reshaping how leadership itself works. The real question isn't whether other CEOs will build the same thing. It's how many already have.
TUTORIAL
Set up Claude Cowork Projects for focused AI work
Recaply: In this tutorial, you will learn how to set up Projects in Claude Cowork to organize tasks, files, and context in one persistent workspace.
Step-by-step:
Download Claude Desktop at claude.com/download and update to the latest version. You'll need a paid plan (Pro, Max, Team, or Enterprise). Cowork isn't on the free tier.
Open Claude Desktop and click the Cowork tab in the sidebar to switch modes. Then go to the Projects panel on the left to see existing projects or start a new one.
Click "New Project" and give it a name. Add a project-level instruction in the Instructions field. Tell Claude what this project is for, your preferred output style, and any context it should always keep in mind.
Go to the Files section and add any documents, links, or context this project needs. Claude will reference these across every task. Files stay local on your computer and aren't shared with Anthropic.
Start a task by describing what you want done, then step away. Come back to find finished work — documents, research, organized files — all scoped to this project's context and memory.
Pro tip: Use the scheduled tasks feature to have Claude run recurring work on its own, like a weekly competitive summary or a daily content digest. Set it once and it runs without any prompting.
PALANTIR & PENTAGON
Pentagon locks Palantir into all US Military forces
Recaply: The Pentagon just designated Palantir's Maven Smart System as an official program of record, locking in long-term funding for the AI weapons-targeting platform across all US military branches.
Key details:
Maven analyzes data from satellites, drones, radars, and sensors to spot targets on the battlefield. The US military has already used it in thousands of strikes against Iran.
The Pentagon holds a $1.3B contract with Palantir for Maven, up from $480M in 2024. Palantir's market cap sits at roughly $360B, with its stock doubling in the past year.
One reported complication: Maven uses Anthropic's Claude AI tool, but the Pentagon has called Anthropic a supply chain risk, creating a tension between its reliance on Maven and its concerns about the underlying AI.
The change is expected to take effect by September 2026. Maven oversight moves to the Pentagon's Chief Digital AI Office within 30 days, per a letter signed by Deputy Secretary Steve Feinberg.
Why it matters: The US military has picked its AI platform. That's bigger than any software deal. One company now controls the AI targeting layer across every branch of the armed forces. The stakes around who builds those models and what guardrails they carry are enormous. The Anthropic issue is just the first preview of harder questions still coming about accountability in military AI.
TOOLS
Trending AI Tools
💻 Google AI Studio - Google's new full-stack vibe coding platform
🗂️ Claude Cowork Projects - Anthropic's Cowork now has Projects
🤖 Kimi K2.5 - Moonshot AI's open-source coding model, now publicly known as the foundation for Cursor's Composer 2
✍️ Write - WordPress's new write capability lets Claude, ChatGPT, and other AI agents draft posts, build pages, and manage site content through natural conversation
NEWS
What Matters in AI Right Now?
OpenAI plans to nearly double its staff from 4,500 to 8,000 by end of 2026. Most new hires will go into product, engineering, research, and sales, the Financial Times reported.
HSBC is looking at cutting around 20,000 jobs, or 10% of its workforce. CEO Georges Elhedery is betting on AI to shrink the bank's middle and back offices.
A North Carolina man pleaded guilty to music streaming fraud using AI. Michael Smith used AI to make hundreds of thousands of songs, then ran bots to stream them billions of times, collecting over $8M from Spotify, Apple Music, and others.
Anthropic launched Projects in Cowork. The feature lets users organize tasks into their own workspaces, each with files, links, instructions, and memory.
WordPress added write tools to its AI agent integration. Claude, ChatGPT, and other AI tools can now draft posts, build pages, manage comments, and organize content directly on users' sites.
The Wall Street Journal reported that AI is changing how Google Search works. AI-generated answers are replacing traditional results, shifting how content creators get traffic.
Elon Musk shared plans for a chip factory he's calling "Terafab," targeting 100 to 200 gigawatts of AI computing per year for Tesla and SpaceX. No timeline was given.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
