The 2-minute brief for AI builders.

👋 Welcome to The 2-minute brief for AI builders.

In today's edition:

• Vercel launched an open-source agent framework plus a secure stack for shipping agents
• Cursor is getting its own code hosting product built for teams and agents
• Lovable opened its MCP server to all users so you can build apps from your AI client
• Anthropic shared what 400K Claude Code sessions reveal about how people are actually using agentic coding

🛠️ Vercel just shipped a full agent stack

Vercel introduced eve, an open-source agent framework it describes as “like Next.js, for agents.” It also rolled out Vercel Connect in public beta and a new package for enterprise apps and agents after saying it already runs 100+ agents in production.

• eve includes building blocks for durable execution, sandboxed compute, approvals, channels, tracing, and evals
• Vercel’s broader Agent Stack bundles AI SDK, AI Gateway, Workflow SDK, Sandbox, Chat SDK, and Vercel Connect
• Vercel Connect lets apps and agents access Slack, GitHub, and other services with short-lived tokens and precise scopes, instead of storing long-lived secrets
• The enterprise package adds identity, access controls, and audit trails for teams shipping agents inside larger companies

What it means: Agent infra is getting bundled into opinionated platforms. If you’re building internal tools or customer-facing agents, this lowers the amount of security and workflow plumbing you need to write yourself. It also signals that “production agent” now means more than model calls—it means auth, isolation, approvals, and logs from day one.

Read more on Vercel →

📦 Cursor is moving into code hosting

Cursor announced Origin, a new code storage and git hosting product for teams and agents. The launch is planned for this fall, with a waitlist open now.

• Cursor says Origin gives teams and agents a place to host, review, and collaborate on code
• The move pushes Cursor beyond coding assistance and into the workflow layer where code gets managed and shipped
• That matters because agent-written code creates new review and merge problems that traditional git tools were not built around
• For builders already deep in Cursor, this points to a tighter loop between writing code, reviewing changes, and coordinating humans with agents

What it means: The coding assistant race is moving up the stack. If your product depends on agentic coding, start paying attention to where code lives and how reviews happen—not just which model writes best. Expect more tools that treat agents like first-class collaborators inside version control.

Read more on Cursor →

🔌 Lovable opened its MCP server to everyone

Lovable said its MCP server is now available to all users and on every plan, letting your AI assistant or editor create, iterate on, and deploy full-stack apps on Lovable without leaving your current client.

• The promise is simple: build without leaving your client of choice
• Lovable says the MCP server lets an AI assistant or editor create, iterate, and deploy full-stack apps directly on its platform
• That means less copy-pasting between chat tools and app builders, and more agent-friendly workflows
• In a separate community post, Lovable highlighted people using Lovable with MeetGranola to build things like group trip planners and revenue forecasting tools

What it means: MCP keeps turning AI products into endpoints that other agents can operate. For founders and indie builders, that opens a practical path to “chat-to-app” workflows inside the tools you already use. If you build on Lovable, this makes it easier to turn your preferred AI client into the front end for app creation.

Read more on Lovable →

📊 Anthropic showed what 400K Claude Code sessions look like in the wild

Anthropic published new research on Claude Code usage, analyzing 400K sessions from October 2025 to April 2026 with a privacy-preserving tool to understand who is using agentic coding, what they do with it, and what drives success.

• More than half of sessions were mainly about writing or repairing code
• Nearly 1 in 5 sessions were about operating software
• Anthropic says the monetary value of the average session, based on comparable freelance marketplace work, grew 27% from October to April
• On its toughest success measure, every occupation was within 7 percentage points of software engineering
• Domain experts were more likely to succeed, but Anthropic says the gap between intermediate and expert users was quite modest

What it means: Agentic coding is spreading beyond engineers, and the work people trust it with is getting more valuable. If you sell AI tools, this is a strong signal to target operators, analysts, and other domain experts—not just developers. If you use coding agents yourself, knowing the domain still helps, but you may not need elite technical depth to get useful results.

Read more on Anthropic →

👋 Thanks for reading. If you liked this roundup, forward it to a friend!

Want to advertise on this newsletter? Click here.

This newsletter was created with the assistance of artificial intelligence. While we strive for accuracy, please note that AI-generated content may occasionally contain errors or biases.