Welcome to the wildest week in AI yet.
From infinite agents to deep fake car shows, from Chrome-based copilots to Claude 4’s arrival—the AI arms race just went nuclear.
Let’s start with Google, who couldn’t even wait for their own conference. The day before I/O 2025, they dropped a mobile app for NotebookLM, making it easier than ever to collect info and listen to notebook podcasts on the go. Then came the leak-turned-launch of Jules, a new AI coding agent that opens PRs, writes tests, and clearly wants to throw down with OpenAI’s Codex.
But the real tease came from DeepMind’s Demis Hassabis, who posted AI-generated video + audio and hinted at Veo 3—a model many suspected would debut the next day. Spoiler: it did.
And speaking of teasers, Flowith AI stunned the community with Flowith Neo, an “infinite agent” promising infinite context, steps, and output. Sleep well, it works while you don’t. If it’s legit, it’s a game changer.
Meanwhile, Microsoft dropped three power moves:
Native MCP support for Windows
VS Code becoming an open-source AI editor
And Grok coming to Azure Foundry in June
Then came Google I/O Day 1, and it was wall-to-wall AI.
Gemini is going full assistant mode in Chrome, drawing from your browsing history and Google Apps to answer contextual questions. Gemini Live will be integrated across your entire Google suite.
Google Meet also got a glow-up with real-time, tone-preserving translation, starting with English and Spanish.
Then came Gemini 2.5 Pro’s “Deep Think Mode”, introducing enhanced reasoning powered by parallel thinking—perfect for coding and logic tasks.
Google also introduced a new Gemini Diffusion model, which iteratively refines outputs instead of predicting text directly. It’s built for math and code, and it’s fast. Then came the big moment: Veo 3, Google’s next-gen video model that includes generative sound, so characters not only move, but talk—from a single prompt.
Partnering with that is Google Flow, a slick new editing + storyboarding tool to string together scenes while maintaining character and visual consistency. And for music creators, Lyria 2 expands the Music AI Sandbox—a real shot across Suno’s bow.
Fashion-forward? Google’s launching a virtual try-on tool in the US: upload a photo, see yourself in different outfits. Experimental now, but this could become mainstream fast.
And on the hardware side?
Android XR brings mixed reality across devices—from glasses to full VR. Google Glasses are back, now with built-in AI.
Samsung’s Project Muhan is the first XR headset, dropping later this year. And then there’s Google Beam, a wild multi-cam setup that turns 2D video calls into 3D Zoom meetings with millimetre-accurate head tracking.
Not to be outdone, OpenAI and Jony Ive made shockwaves with a 10-minute reveal. Ive’s hardware startup IO is merging with OpenAI to develop a family of AI devices. Altman called the prototype “the coolest piece of tech the world has ever seen.” Game. On.
Google then launched Stitch, a generative UI tool that designs interfaces and exports working code—paired with Jules, it made coding the hot battleground of the week. So much demand, they had to throttle access.
Still not enough? Claude 4 is here.
Anthropic’s latest flagship comes in Sonnet and Opus models and claims top-tier performance in math and code. They’re betting big on agentic coding, promising to challenge the very best developer models out there.
Perplexity followed up with plans for browser-based task scheduling, best experienced in their upcoming Comet browser. Seamless execution of tasks straight from chat? Yes, please.
And the bots? Oh, they’re warming up.
Mech Combat Arena kicks off this weekend in China, and robot fighters are looking clumsy—but that might just be strategy. Don’t show all your weapons before battle day.
Back to Google to close out the week:
Veo 3 delivered some mind-blowing video generations, with sound, characters, interviews—even fake car shows that look 100% real. This isn’t just deepfake territory anymore—it’s world-building in a prompt.