👋 Hello hello,
A rocket company just bought a code editor. SpaceX is paying around $60 billion for Cursor, which means the same people landing boosters on barges now own the tool a huge chunk of developers open every morning. The logic gets less strange once you realize the real prize is compute, and SpaceX has plenty of it.
While that deal ate the headlines, a Chinese lab handed everyone a frontier-grade model for free, and Microsoft flipped a switch that turns its AI from a helpful intern into something that completes your work.
Let's get into it.
🔥🔥🔥 Three Curated AI Updates
On June 16, SpaceX agreed to acquire Anysphere, the company behind the Cursor AI code editor, in a deal valued near $60 billion. CEO Michael Truell said he's excited to partner with the SpaceX team to scale up Composer, Cursor's in-house coding model.
Quick context if you don't live in developer tools: Cursor is the editor a lot of engineers now use to write and ship code with AI doing most of the typing. Composer is the model under the hood.
Fontier models are bottlenecked by compute, and tying Cursor to SpaceX gives it the capital and infrastructure to train bigger models without burning its own runway. It also shows how valuable the "where developers actually work" surface has become, valuable enough for a space company to pay eleven figures for it.
Z.ai (formerly Zhipu AI) released GLM-5.2, a frontier-class open model built for coding and long, multi-step agent tasks. It ships with a 1M-token context window, two reasoning levels (a max mode that pushes hard and a high mode tuned for token efficiency), and an MIT license, all at the same API price as GLM-5.1.
Context: "open weights" means you can download the model and run it on your own machines instead of only renting it through an API, and the MIT license means almost no strings attached.
A free, frontier-level model you can self-host is a big deal for anyone worried about cost or about a provider yanking access. One caveat worth knowing: Z.ai's performance claims haven't been independently benchmarked yet, so treat the "beats the big labs" framing as a claim for now.
Weights: huggingface.co/zai-org/GLM-5.2
Microsoft made Copilot Cowork generally available everywhere, now with multi-model support. It's an agentic system for complex, long-running tasks: you describe the outcome, and it runs the whole thing end to end, then hands back a finished result instead of a draft.
Name check, since it gets confusing: this is Microsoft's Copilot Cowork inside Microsoft 365, which is a different product from Anthropic's Claude Cowork. "Multi-model" means it can run on different underlying models depending on the task.
Why it matters: this is Microsoft moving Copilot past "draft this email" into "go run this multi-step project." It needs a Microsoft 365 Copilot license and bills by usage, so the cost scales with how hard you work it.
🔥🔥 Two Tools To Know
1. 🔎 Exa Agent
An API built for serious web research. You hand it a task, like enriching a list of companies or building a big dataset, and it orchestrates a mix of cheaper models to finish the job at less than half the cost of running GPT-5.5 or Opus on the same work. It's developer-facing, so this one's for folks comfortable wiring things into their own stack.
2. 🎙️ Invoko
A Mac assistant you talk to. Hold a key, speak from the notch, and it reads whatever's on your screen to draft a reply, summarize a page, rewrite a selection, or take an approved action across your apps. It remembers context between sessions, and your voice never gets stored on their servers. Best for anyone who lives on a Mac and would rather speak than type through the busywork. Apple Silicon only for now, in beta, with a Windows waitlist.
🔥 Things You Didn't Know You Could Do With AI
Someone built a McDonald's vs KFC fight video entirely with AI, and it's exactly the kind of dumb-fun clip that racks up views. You can make the same thing for your own brand. Here's the simple version:
Pick a concept with built-in tension, like two rival mascots squaring off or your product swooping in to save the day.
Write it as a shot list, one sentence per shot, covering camera angle, action, and mood.
Generate each shot in an AI video tool (Veo, Sora, Kling, and Runway all work), keeping clips to 5 to 8 seconds.
Stitch the clips together, layer in sound effects and music, and cut anything that drags.
Post it where short video lives and let the comments do your marketing for you.
Here's the video that inspired this workflow:
Don't forget to rate today's post
Until next time,
Team @PracticalyAI
