Google’s latest showcase wasn’t about posting the highest scores. It was about making AI feel useful in daily life. After a week of watching demos, talking with teams, and tracking reactions, my take is clear: Google is right to chase usefulness—but it’s playing with fire on search, privacy, and product sprawl.
From Benchmarks to Everyday Use
Gemini 3.5 Flash set the tone. It trades a bit of raw skill for speed and low cost, landing near top-tier models in many tasks while running much faster and cheaper. For most people, that matters more than beating a test by a point or two.
The showstopper, though, was Gemini Omni. It turns prompts and mixed inputs into short, knowledge-grounded videos, edits your clips, and keeps character consistency. The team pitched it as “anything in, video out” for now, with more formats on deck.
“Right now you can put any modality into the model as input, images, videos, text, audio, and generate video. But in the future… image outputs and audio outputs as well.”
Early results are messy in spots—odd angles, extra objects—but the direction is obvious. Natural-language video creation is becoming real enough for teaching, marketing, and fast explainers.
Then came Gemini Spark, an agent that operates across your Google life—Gmail, Calendar, Drive—and runs in the cloud even when your laptop sleeps. The promise is action, not chat.
“You might say every morning send me a motivating email. Or summarize the news and write it in a doc… it can actually do things for you that you don’t need to watch every time.” — Karee, CTO of Google DeepMind
Google stressed guardrails. Agents ask before touching the outside world. That’s smart design—and it needs to stick.
Where The Shine Fades
One choice towers over the rest: search. Google is moving to longer prompts, code and visuals in results, and AI summaries that answer without a click. That may please users in the moment. But it risks gutting the open web that supplies those answers.
Creators rely on visits. If AI summaries crowd out links, fewer sites will survive to write the next guide or review. YouTube’s new “Ask” feature points the same way, answering from videos without a view. If Google doesn’t share attention—and value—the source well runs dry.
Two other concerns stood out during the week.
- Product confusion: Too many overlapping tools for code, design, and video. It’s hard to know where to start.
- Agent trust: Spark’s access is powerful. Users must stay in charge of spending, messaging, and data flow—always.
- Shopping control: Universal Cart and agent payments sound handy, but missteps with money or bad recommendations will erode trust fast.
To be fair, Google also scored real wins. SynthID watermarks for AI media—with rivals like OpenAI and 11 Labs on board—could help slow deepfakes. Glasses tied to Gemini may finally reach buyers this fall. And Project Genie’s Street View scenes show how quickly playful ideas are turning interactive.
The Stakes For Everyone Else
Outside Mountain View, the week brought more signals. OpenAI tied ChatGPT to bank data for personal finance coaching. That convenience collides with ad ambitions and data sensitivity. I don’t want a chatbot mixing spending history with sponsor influence.
Spotify and Universal struck a path for licensed AI covers and remixes. That’s the first serious attempt to pay rights holders for AI-made music at scale. It’s overdue.
And yes, Andrej Karpathy joining Anthropic matters. Talent reshapes the pecking order as much as model releases do.
My Bottom Line
Usefulness is the right north star—but not at the cost of a healthy web, clear control, and user trust. Google can win by doing three things: keep link equity visible in search, make agents opt-in at every risky edge, and merge overlapping tools into a cleaner path.
Readers should pressure platforms to show sources, share traffic, and ask permission before agents act. Vote with use. Reward tools that respect creators and keep you in charge.
We don’t need perfect demos. We need honest trade-offs, stable rules, and a web that still pays the people who make it worth visiting.
Frequently Asked Questions
Q: What makes Gemini Omni different from typical text-to-video tools?
It doesn’t just turn a prompt into clips. It edits your footage, mixes inputs like images and audio, and aims to keep character details consistent across shots.
Q: Why is the new search approach drawing pushback?
AI summaries can answer without sending visitors to the sites that did the work. Less traffic means fewer creators, which weakens the source material for future answers.
Q: Are Google’s agents safe to let loose on email and calendars?
They’re designed to ask before acting outside your accounts, like emailing others or spending. Keep permissions tight and review logs until you trust the behavior.
Q: Is Gemini 3.5 Flash good enough for coding work?
It’s fast and inexpensive, and it performs near higher-end models in many tasks. For complex codebases, you may still want to escalate to a top model as needed.
Q: How will AI-made music on Spotify pay artists?
The licensing deal enables revenue sharing on fan-made covers and remixes. Streams of AI versions can credit and compensate the original artists and writers.
