👋 Hello hello,
Somewhere in a bedroom studio right now, someone is remixing their favorite song and sitting on the uneasy question of whether it'll get them sued. As of this week, that answer changed.Spotify and UMG announced a licensing deal that makes fan remixes legal and monetized, for both the creator and the original artist.
Microsoft shipped AI agents that can operate any software on screen the way a human would, something enterprise IT teams have been asking for since forever. And OpenAI's 2026 election safeguard plan landed this week too, with a live vote count partnership with the Associated Press backing it up.
Let's get into it.
🔥🔥🔥 Three Curated AI Updates
Spotify and Universal Music Group announced a landmark licensing deal that gives fans the ability to create AI-powered covers and remixes of songs from participating artists and songwriters. Artists earn income every time a fan remix is created with their music, and the whole framework is built on what Spotify co-CEO Alex Norström called "consent, credit, and compensation."
The tool launches as a paid add-on for Spotify Premium users. Fan remix culture has always existed, from SoundCloud uploads to bedroom mashups, sitting in a grey legal zone for decades. This is the first time a major platform has moved it into a structure where everyone in the chain actually gets paid. For generative AI and the music industry, it is a genuine template for how this is supposed to work.
Every company has software with no API: old vendor portals, legacy CRMs, internal tools from a decade ago that still run core operations. Automating anything in those systems historically meant a human had to sit there and click through screens. Microsoft just changed that.
Copilot Studio's computer-use agents hit general availability on May 13, making Microsoft the first major platform to ship this capability at production scale. The agents can see any graphical interface and interact with it the way a person would, clicking buttons, filling fields, and selecting menus. The GA release adds model choice between Claude Sonnet 4.5 and OpenAI CUA, Azure Key Vault credential storage, Microsoft Purview audit logging for enterprise compliance, and Windows 365 Cloud PC pool support for isolated execution. Anthropic and Google are both still in preview. Pricing runs at roughly $0.04 per step.
2026 is the second major global election year since generative AI went mainstream, and OpenAI laid out its full game plan this week.
Five pillars:
surfacing reliable voting information,
supporting cybersecurity defenders,
watermarking AI-generated content via a new SynthID partnership,
enforcing usage policies against election interference, and
increasing transparency across the board.
The most concrete piece is a new deal with the Associated Press to surface live US vote counts in ChatGPT starting this fall, extending through the 2028 cycle. OpenAI is also making its Codex Security and Trusted Access for Cyber tools available to registered voting system manufacturers in the US, and partnering with Democracy Works to show accurate voter registration information directly in ChatGPT.
AI companies took real heat for misleading political content in 2024. This is the most structured response yet.
🔥🔥 Two AI Resources Worth Checking Out
1. 🍜 EPICURE by Jakub Radzikowski and Josef Chen
EPICURE is a research model that compressed all of human cooking into roughly 2 megabytes. The team trained it on 4.14 million recipes across 7 languages, mapping 1,790 ingredients across 300 dimensions to mathematically represent flavor, culture, and pairings. It can calculate something like the directional "vector between Chinese and Ethiopian cuisine," which turns out to be surprisingly interpretable. For anyone building food apps, recipe tools, or just interested in where AI and culinary science overlap, this is the kind of foundational resource that does not come around often.
Higgsfield is an AI video generation platform, and its plugins for Premiere Pro and After Effects just went live. From inside your editing timeline, you can now generate images, videos, and transitions; reframe aspect ratios without cropping; remove backgrounds from footage; sketch directly on a clip to describe the edit you want; and export upscaled to 4K. The full AI creative suite now lives where video editors already work, so there is no switching between tabs and tools.
🔥Things You Didn't Know You Could Do With AI
Bilawal Sidhu, a former Google product manager, took an aerial screenshot of Austin from Google Maps, drew a squiggly line over it as a camera path, handed the image to Gemini Omni Flash, and got back realistic drone POV footage flying along exactly that route. He started with 3D scan camera poses as a reference, realized a freehand scribble on a map screenshot worked just as well, and shared the results publicly.
What this demo shows is that a rough spatial sketch is enough to direct the model like a virtual drone operator. You are giving it a creative brief drawn with a marker, and it figures out the rest.
Here is how to try it yourself:
Take a screenshot of your location from Google Maps or Google Earth in aerial view.
Open any drawing tool and sketch a flight path over the image, or sketch it on paper, photograph it, and crop it down.
Open Gemini Omni Flash in the Gemini app or Google AI Studio and upload your annotated screenshot.
Prompt: "Using this sketched path as a camera direction, generate drone POV footage flying along this route over [city name]."
Prompt a second time to clean up: "Regenerate the same footage but remove all the sketched lines from the video."
If you want to see exactly how this looks in action, watch the full video here.
💬 Quick poll: If you had access today, which would you try first: Spotify's fan remix tool, Microsoft's computer-use agent, or Gemini Omni's video generation?
Did you find today's post useful?
Until next time,
Team @PracticalyAI
