Reports

Thu, May 8 - Google Launches Gemini 2.5 Pro with Video-to-App Tools

Hak

08 May 2025 — 4 min read

It's Thursday, May 8, 2025 and you're reading the Agentive Daily Report.

Busy People's Section

Google's Gemini 2.5 Pro launches with enhanced coding abilities and a groundbreaking capability to generate applications directly from video input.

OpenAI reverses course on its corporate structure, confirming its nonprofit will retain control while planning to reduce its revenue sharing with Microsoft.

Netflix updates its TV app with AI-powered search, smarter recommendations, and a TikTok-style vertical video feed.

Pinterest enhances its visual search with new tools for narrowing results and exploring styles, focusing first on women's fashion.

Lightricks releases LTX Video 13B, an open-source AI video model that generates high-resolution video up to 30 times faster than similar models.

Nvidia releases Parakeet-TDT-0.6B-V2, a fully open-source speech recognition model that can transcribe an hour of audio in one second.

Today's Top Stories

Google Unveils Gemini 2.5 Pro with Video-to-App Generation

Google has released Gemini 2.5 Pro I/O Edition, which has emerged as the leader in AI coding, claiming the top spot on the WebDev Arena leaderboard. The model introduces groundbreaking video-to-code capabilities, allowing developers to generate complete applications directly from video input. These advancements enable use cases like converting YouTube tutorials into interactive learning platforms.

Beyond video processing, the update enhances web app generation, UI design, and automates web project feature integration. Available through Gemini API and Vertex AI with no price increase from previous versions, Google is positioning this release to cement its place in the competitive AI development landscape, particularly against rivals like GitHub Copilot and Claude.

OpenAI Restructures and Takes on Microsoft

In a strategic pivot, OpenAI has reversed its decision to transition into a for-profit entity, confirming that its nonprofit will retain control. The company plans to convert its for-profit arm into a Delaware public benefit corporation under nonprofit oversight, replacing the capped-profit model with standard shares while maintaining regulatory assurances about control.

Financially, OpenAI is looking to reduce the approximately 20% revenue share it currently pays Microsoft (its $13.75B investor) to around 10% by 2030. The company is also acquiring coding startup Windsurf for roughly $3 billion, its largest acquisition to date. This restructuring creates a new governance landscape that balances commercial interests with OpenAI's mission, though observers note it may complicate any future IPO plans.

Netflix Launches AI-Enhanced Interface and TikTok-Style Feed

Netflix is rolling out a significant update to its TV app that incorporates AI-powered search, smarter content suggestions, and a cleaner design. The update's standout feature is a vertical short video feed similar to TikTok or YouTube Shorts that will showcase clips from trending shows and movies.

The platform is leveraging generative AI tools to improve search and recommendations, helping users discover content more intuitively. This move demonstrates Netflix's strategy to stay competitive in an environment where short-form video is increasingly popular and AI is reshaping content discovery. The worldwide update will deploy in the coming weeks on Android and iOS, reflecting how even established streaming platforms are embracing AI to enhance user experience.

Fast Forward

Lightricks Open-Source Video Model: Lightricks has launched LTX Video 13B, an open-source AI model that generates high-resolution video up to 30x faster than competitors, featuring multi-scale rendering and enhanced creator controls while being trained on licensed data for commercial safety.
Pinterest Visual Search Upgrade: Pinterest has enhanced its visual search capabilities with new tools that help users narrow results and explore styles by category, launching first in women's fashion across select regions to make image-based discovery more precise and intuitive.
AI Art Authentication: Artificial intelligence detected unusual features in Raphael's famous painting "Madonna della Rosa," suggesting that Saint Joseph's face was likely painted by a student rather than the master himself, demonstrating how AI can assist art historians in attribution work.
Cognition KEVIN-32B Release: Cognition has launched KEVIN-32B, a specialised model using reinforcement learning for multi-turn code generation that outperforms existing models in developing CUDA kernels, optimising performance through intermediate feedback and efficient reward distribution.
Reddit Anti-Bot Measures: Reddit is tightening its verification policies to keep human-like AI bots off the platform following a controversial experiment where researchers deployed AI bots in discussion forums without authorization, highlighting growing concerns about AI manipulation in online spaces.

New Tools Discovered

Parakeet-TDT-0.6B-V2: Nvidia's new open-source speech recognition model that tops performance benchmarks while being able to transcribe an hour of audio in just one second, including punctuation and word-level timestamps.
Clippy: A nostalgic tool that allows users to run various language models locally on their computer with a 1990s interface reminiscent of Microsoft's classic assistant.
UI Builder + Mockup tool: A Chrome extension that lets designers sketch UI mockups directly on live websites, streamlining the process of reimagining existing interfaces.
ByteNite: A distributed computing platform designed specifically for AI applications that goes beyond traditional serverless architectures.
Miniflow: A user-friendly platform that enables non-technical users to create and automate AI workflows without coding knowledge.

Discover more tools at Agentive.Directory

That's a wrap for today! Thank you for reading this report.

Have thoughts on today's edition? Hit reply and let us know what you're thinking. Or if you've discovered a cool AI tool we should feature, drop us a line.

Until tomorrow,
Hak from Agentive.Studio