Google doesn’t just want answers. It wants agency.
Gemini Spark is live, or at least announced, and it sounds like the digital butler everyone was dreading and hoping for in equal measure. It’s Google’s shot across the bow at OpenClaw. While your computer naps, Gemini stays awake. It reads data. It sends emails. It spends money. You didn’t even have to click a button. Is this helpful or terrifying? Depends on how much trust you put in code to handle your rent.
Gemini Spark Is Google’s Response to OpenAgents’s 24/7 Agent
This ties into the bigger Google I/O 2029 spectacle. The tech giant isn’t just polishing its search engine. It is rewiring it. New Gemini models are sharper, obviously, but the search itself is becoming a background process. Add some smart glasses dropping this fall—glass that looks like glass, presumably—and you have an ecosystem that watches, listens, and acts without you raising your hand.
Then there is Flow.
Google overhauled its AI creation suite, and suddenly “deepfake yourself” is a feature, not a bug. Avatars generate selfie videos of you. It looks like you. It says things you might not say. The line between authenticity and generation has dissolved into a blurry pixel mess. Why bother with truth when a server can simulate a better version of you for a meeting?
Android 17 follows a similar logic. Coming this summer. Widgets generated by AI. Chrome on Android asking Gemini to finish your bookings because, presumably, clicking is too slow. We are outsourcing the clicking. The scrolling. The minor decisions.
Meanwhile, the people in charge are confused.
Demis Hassabis, the head of Google DeepMind, thinks AI layoffs are dumb. WIRED caught up with him, and he argues that AI boosts productivity so we should do more work. Not less. The idea is that automation liberates humans to tackle higher-value tasks instead of shrinking headcounts. Optimistic. Maybe too optimistic. Mira Murati agrees, mostly. The former OpenAI CTO and Thinking Machines Lab founder wants AI that collaborates, not one that fires. Keep humans in the loop, she says. But who is steering the ship if the AI is tying the knots?
Demis Hassabis Thinks AI Job Guts Are Dumb
On the other side of the industry, OpenAI is tweaking the details. ChatGPT’s Image Model 2.0 arrived, and our tests show it renders text better. Less garbled script, more clarity. But it still chokes on anything that isn’t English. A familiar flaw.
OpenAI’s Codex agent has weirder rules.
The instructions are strict. No goblins. No gremlins. No raccoons. The code-writing AI is banned from hallucinating mythical creatures or real animals unless it is strictly relevant. “Shut up about the raccoon,” essentially. It seems absurd until you consider how easily these models drift into narrative territory.
The Bloomberg Terminal is getting a chatbot overlay now, too. Traders don’t necessarily like it. They don’t want a chatty assistant; they want data. But the CTO insists the change is inevitable. Like it or not.
The Bloomberg Terminal Is Getting an AI Facelift, Like It or Not
There are deals to be had if you don’t care about the philosophy of automation. WIRED is running promos for Google Workspace in May. Up to 14 percent off. Three months. Starter, Standard, or Plus plans. Buy now, save money, get automated later.
We are moving fast. We are also tired of the branding.
Anthropic announced their agents have “dreams.” They process “memories.” Someone please stop naming AI features after human biological functions. It’s creepy. It implies sentience where there is only pattern matching. Can we have boring names again? Like “Data Processor 4.” Or just “Code.”
The future is agentic. It spends our money while we sleep. It deepfakes our faces for Tuesday morning calls. And we keep asking for faster widgets.
Who is actually in control?
The user interface disappears. The agent remains.
