Cursor ships web apps to Vercel

Cursor links app creation to deployment

Cursor can now create a web app, store its code with Origin, and deploy it to Vercel. That is a much shorter path from a build session to a live URL.

• Create new web apps inside Cursor.
• Store the resulting code with Origin.
• Deploy the app to Vercel from the same workflow.

What it means Use this flow to test small product ideas faster, but keep a review step before deploying anything that handles customers, payments, or sensitive data.

Source: cursor_ai

Gemini adds more control to AI video

Google introduced Gemini Omni 1.1 Flash, a multimodal model for video generation and editing that brings more controls into the builder workflow.

• Draft videos faster at 360p before committing to a higher-resolution result.
• Set the first and last frame to steer a scene's opening and ending.
• Upscale output to 4K.
• Extend scenes for longer storytelling and build directly in Google AI Studio.

What it means Prototype video ads, product demos, and social assets at 360p, then use frame controls and 4K upscaling only on the versions worth shipping.

Source: GoogleAI

Cohere targets the document-workflow bill

Cohere launched Parse 5, a document parsing model aimed at high-volume enterprise jobs where extracting usable data from messy files can get expensive.

• It recognizes text, tables, forms, and images, producing machine-readable documents with bounding boxes.
• Cohere lists use cases including structured-data extraction, RAG systems, and automated document processing.
• The company prices it at $1.50 per 1,000 pages.
• Parse 5 is available through the Cohere API, Model Vault, AWS SageMaker, Microsoft Foundry, and a free Hugging Face Space.

What it means Price-test Parse 5 for ingestion-heavy RAG or back-office automation, especially when tables and forms make plain text extraction fall apart.

Source: cohere

Anthropic starts a hardware standard preview

Anthropic began the first research-preview phase of Model Hardware Standard, a proposed standard for AI agents to safely operate physical equipment in science and advanced manufacturing.

• MHS aims to replace days or weeks of bespoke hardware integration with hours or minutes of work.
• It provides a discoverable device interface and a way for agents to operate equipment safely.
• Early tests included a drug-discovery experiment at Genentech with real-time error handling.
• Anthropic says it plans to involve science, robotics, electronics, and manufacturing stakeholders before opening the standard to the open-source community.

What it means If you build agents around labs, factories, boards, or cameras, follow MHS now because a shared device interface could shrink future integration work.

Source: AnthropicAI

Replit puts model selection on autopilot

Replit launched Intelligent Model Routing, which selects a model for each task so builders do not have to manually trade off output quality, speed, and cost.

• Replit says the routing can deliver top-tier output for up to 65% less cost.
• In Agent, the “auto” toggle matches the model to each prompt by default.
• The product removes manual switching between modes, models, and effort levels.
• Enterprise routing selects an optimal compliant model for each task, according to Replit.

What it means Turn on automatic routing for mixed build workloads, then compare your real spend and output quality against a fixed-model setup before making it your default.

Source: Replit

Open source models are about to change EVERYTHING

A video on how open-source models may affect AI builders.