AI Daily Dev — May 19, 2026

Gemini Omni Lands at I/O as Google's First Unified Video, Audio, and Image Model

Single model generates synchronized video, audio, voice, and text in one pass — replacing the Veo 3.1 + Nano Banana split pipeline.
Chat-based editing: 'swap the red car for a black one' or 'remove the watermark' rewrites only the affected frames, keeping the rest pixel-stable.
Inherits Gemini's long context — characters keep their faces, outfits, and props across scenes of a short film.
Rolls out in Flash and Pro tiers across the Gemini app and AI Studio; the UI leak that surfaced two weeks ago ('Powered by Omni' next to codename Toucan) is now live.

models blog.google

New top-tier Gemini debuts in AI Mode in Search on day one — first time Google has shipped a frontier model into Search at launch.
Pre-keynote benchmarks circulating cite ~84.6% on ARC-AGI-2 and meaningful jumps on multimodal reasoning, but still trail Anthropic's April-7 Mythos and OpenAI's GPT-5.5 on the hardest agentic-coding evals.
Available via the Gemini app, AI Studio, and the Gemini API; Antigravity IDE picks it up as the default agent model.
Demis Hassabis framed it as the model that unifies all Gemini capabilities — single-model multimodality, deeper reasoning, lower latency at the Flash tier.

models blog.google

Background agent that polls your connected apps, learns your patterns, and acts proactively — the surface that replaces Project Mariner (shut down May 4).
Cleans up Gmail, drafts meeting briefs from across your docs, and assembles a personal news digest without prompting.
Asks permission before purchases or sensitive sends, but Google still flags it as 'experimental' and notes risk of unintended actions.
Compared head-to-head with Claude Cowork in coverage; ships inside the Gemini app for paid tiers first.

tools 9to5google.com

Two tiers shown: a display-free AI pair with camera + mics + speakers for hands-free Gemini, and an in-lens display variant for navigation and live translation.
Samsung's 'Jinju' reference design: 12MP Sony IMX681, Snapdragon AR1, 155 mAh battery, photochromic lenses, ~50g frame.
First Android XR partners include XREAL's optical-see-through Project Aura headset (70° FoV, CES Innovation Award winner).
Consumer pricing teased at $379–$499 for the audio-only tier; ships into a smart-glasses category facing its first real privacy backlash.

industry androidauthority.com

New laptop OS merges Android and ChromeOS into a single Gemini-native desktop — bottom dock, virtual desktops, Quick Settings, built-in Link to iOS.
Acer, ASUS, Dell, HP, and Lenovo signed on as launch hardware partners; first Googlebooks ship fall 2026.
Hands-on leaks showed an ARM-first stack with Pixel-style integrations; the official product name is still under wraps ('Aluminium' is the internal codename).
Chromebook line is being wound down — the Pixelbook Go is the last ChromeOS device Google will sell directly.

industry thenextweb.com

Imbad0202/academic-research-skills: full research→write→review→revise→finalize pipeline as a Claude Code skill bundle, +1.4K stars today and ~12.5K total.
Author's stance: 'AI is a copilot, not the pilot' — the skills handle reference-hunting, citation formatting, data verification, and consistency checks; humans keep the thesis.
Five of GitHub Trending's top 20 this week are skill registries — mattpocock/skills, tech-leads-club/agent-skills, K-Dense-AI/scientific-agent-skills, and HKUDS/CLI-Anything (now 36.9K stars) round it out.
Signal: the agent-tooling community is converging on shareable, composable skill packs instead of monolithic agents.

open-source github.com