Three Gemini tools collapsed into one prompt today

I have three Gemini surfaces in my daily stack right now. Pro for meeting transcription. 3 Pro Preview for English-to-German translation when I am preparing Adaptig keynote material. Nano Banana Pro for slide images and case-study illustrations through my deck-design skill. Three different model IDs. Three different API calls. Three different mental models for "which one do I reach for."

This morning at I/O 2026 Google shipped Gemini Omni, and the pitch is: stop reaching for three things. One model, one prompt, multimodal in and out, with Veo for video, Nano Banana for images, and Genie's world-simulation work folded into a single surface that holds character, physics, and prior edits consistent across turns.

I have not run Omni in production yet. Omni Flash rolled to AI Plus/Pro/Ultra subscribers in the Gemini app and Google Flow today, with YouTube Shorts integration this week and API access "in a few weeks." So this is a directional read, not a battle-tested one. But the directional read is significant enough to write down on day one, because the consolidation argument lands.

Here is what changes, concretely, in my workflow if Omni delivers what Google demoed:

My deck-design skill currently calls gemini-3-pro-image-preview for branded slide assets. That works well — Nano Banana Pro is the one image model that holds Adaptig's burgundy plus Helvetica without drifting. If Omni replaces that call with a unified surface that also handles short workshop intro videos with consistent characters across cuts, my video evaluation backlog — Kling 3 versus MiniMax Hailuo versus Veo 3.1 — collapses by a column. Not because Kling or Hailuo got worse, but because the "stays inside the same vendor's logic" property becomes more valuable than I had been pricing it.

The conversational editing claim is the one I am skeptical about until I see it survive a real client brief. "Change the background, keep the character, redo the lighting, push the camera angle right" is the kind of demo that always works on the keynote stage and falls apart when the brief is "make the character look 8% more like our brand ambassador without making it look like the brand ambassador." Persistence-across-turns has been the consistent failure mode of every multimodal model I have tested in the past 18 months. If Omni actually holds, that is the unlock.

Gemini 3 Deep Think also got upgraded today. 48.4% on Humanity's Last Exam without tools, 84.6% on ARC-AGI-2, gold-medal level on the 2025 IMO and IPhO. Those numbers matter to research labs more than to my training rooms. But they matter to my pitch deck in one specific way: when corporate L&D buyers ask "is this AI thing actually getting smarter or are we at a plateau," I have a sharper answer than I had yesterday. The plateau narrative is louder than the reality, and Deep Think on a benchmark designed to test the absolute limits of frontier reasoning is the kind of receipt I can point to in a workshop opener.

The consolidation thesis is the one to watch, though. If Omni works as advertised, the right way to teach corporate AI adoption next quarter is not "here are five tools for five jobs" but "here is the one prompt surface and these are the ways it composes." That is a different workshop. It is also a workshop more clients can absorb in a half-day.

I will write back after I have actually run Omni against a real brief. Probably the next batch of Adaptig case-study illustrations is the right pilot — same prompt set I have been giving Nano Banana for months, same brand-constraint requirements, see what survives.

Day one read. Not a finished pattern. Just the day the toolbox got smaller.

Connect with me on LinkedIn.