But more piles of fascinating research are only useful to Google if they generate that most important of outputs: profit. Most customers generally aren’t yet willing to pay for AI features directly, so the company may be looking to sell ads in the Gemini app. That’s a classic strategy for Google, of course, one that long ago spread to the rest of Silicon Valley: Give us your data, your time, and your attention, check the box on our terms of service that releases us from liability, and we won’t charge you a dime for this cool tool we built.
For now, according to data from Sensor Tower, OpenAI’s estimated 600 million all-time global app installs for ChatGPT dwarf Google’s 140 million for the Gemini app. And there are plenty of other chatbots in this AI race too—Claude, Copilot, Grok, DeepSeek, Llama, Perplexity—many of them backed by Google’s biggest and best-funded competitors (or, in the case of Claude, Google itself). The entire industry, not just Google, struggles with the fact that generative AI systems have required billions of dollars in investment, so far unrecouped, and huge amounts of energy, enough to extend the lives of decades-old coal plants and nuclear reactors. Companies insist that efficiencies are adding up every day. They also hope to drive down errors to the point of winning over more users. But no one has truly figured out how to generate a reliable return or spare the climate.
And Google faces one challenge that its competitors don’t: In the coming years, up to a quarter of its search ad revenue could be lost to antitrust judgments, according to JP Morgan analyst Doug Anmuth. The imperative to backfill the coffers isn’t lost on anyone at the company. Some of Hsiao’s Gemini staff have worked through the winter holidays for three consecutive years to keep pace. Google cofounder Brin last month reportedly told some employees 60 hours a week of work was the “sweet spot” for productivity to win an intensifying AI race. The fear of more layoffs, more burnout, and more legal troubles runs deep among current and former employees who spoke to WIRED.
One Google researcher and a high-ranking colleague say the pervasive feeling is unease. Generative AI clearly is helpful. Even governments that are prone to regulating big tech, such as France’s, are warming up to the technology’s lofty promises. Inside Google DeepMind and during public talks, Hassabis hasn’t relented an inch from his goal of creating artificial general intelligence, a system capable of human-level cognition across a range of tasks. He spends occasional weekends walking around London with his Astra prototype, getting a taste of a future in which the entire physical world, from that Thames duck over there to this Georgian manor over here, is searchable. But AGI will require systems to get better at reasoning, planning, and taking charge.
In January, OpenAI took a step toward that future by letting the public in on another experiment: its long-awaited Operator service, a so-called agentic AI that can act well beyond the chatbot window. Operator can click and type on websites just as a person would to execute chores like booking a trip or filling out a form. For the moment, it performs these tasks much more slowly and cautiously than a human would, and at a steep cost for its unreliability (available as part of a $200 monthly plan). Google, naturally, is working to bring agentic features to its coming models too. Where the current Gemini can help you develop a meal plan, the next one will place your ingredients in an online shopping cart. Maybe the one after that will give you real-time feedback on your onion-chopping technique.
As always, moving quickly may mean gaffing often. In late January, before the Super Bowl, Google released an ad in which Gemini was caught in a slipup even more laughably wrong than Bard’s telescope mistake: It estimated that half or more of all the cheese consumed on Earth is gouda. As Gemini grows from a sometimes-credible facts machine to an intimate part of human lives—life coach, all-seeing assistant—Pichai says that Google is proceeding cautiously. Back on top at last, though, he and the other Google executives may never want to get caught from behind again. The race goes on.
Updated 3/21/2025, 4 PM EDT: Wired has clarified the context of a quote attributed to Pandu Nayak.
Let us know what you think about this article. Submit a letter to the editor at [email protected].
.jpg "Esoteric Programming Languages Are Fun—Until They Kill the Joke")
