What a great idea I had for the first Plaintext of 2025. After following the frantic competition between OpenAI, Google, Meta, and Anthropic to churn out brainier and deeper “frontier” foundation models, I settled on a thesis about what’s ahead: In the new year, those mighty trailblazers will consume billions of dollars, countless gigawatts, and all the silicon Nvidia can muster in their pursuit of AGI. We’ll be bombarded by press releases boasting advanced reasoning, more tokens, and maybe even assurances that their models won’t make up crazy facts.
But people are tired of hearing about how AI is transformational and seeing few transformations to their day-to-day existence. Getting an AI summary of Google search results or having Facebook ask if you want to pose a follow-up question on a post doesn’t make you a traveler to the neo-human future. That could begin to change. In ’25 the most interesting AI steeplechase will involve innovators who set about making the models useful to a wider audience.
You didn’t read that take from me in the first week of January because I felt compelled to address topics related to the newsworthy nexus between tech and Trump. In the meantime, DeepSeek happened. This is the Chinese AI model that matched some of the capabilities of the flagship creations of OpenAI and others, allegedly at a fraction of the training costs. The lords of giant AI now insist that building ever bigger models is more critical than ever to maintain US primacy, but DeepSeek lowered the barriers for entry into the AI market. Some pundits even opined that LLMs would become commodities, albeit high-value ones. If that’s the case, my thesis—that the most interesting race this year would be between applications that brought AI to a wider audience— has already been vindicated. Before I published it!
I do think the situation is fairly nuanced. The billions of dollars that AI leaders plan to spend on bigger models may indeed trigger earth-shattering leaps in the technology, though the economics of centibillion-dollar AI investments remain fuzzy. But I’m more confident than ever that in 2025 we’ll be seeing a scramble to produce apps that make even skeptics admit that generative AI is at least as big a deal as smartphones.
Steve Jang, a VC who has a lot of skin in the AI game (Perplexity AI, Particle, and—oops—Humane) agrees. DeepSeek is accelerating, he says, “a commoditization of the extremely high-value LLM model lab world.” He provides some recent historical context: Soon after the first consumer transformer-based models like ChatGPT appeared in 2022, those trying to provide use cases for actual people concocted fast-and-dirty apps on top of the LLMs. In 2023, he says, “AI wrappers” dominated. But last year saw the rise of a countermovement, one where startups attempted to go much deeper to create amazing products. “There was this argument, ‘Are you a thin wrapper around AI, or are you actually a substantial product in your own right?’” Jang explains. “‘Are you doing something truly unique while using at your core these AI models?’”
That question has been answered: Wrappers are no longer the industry delight. Just as the iPhone went into overdrive when the ecosystem shifted from clunky web apps to powerful native apps, the AI market winners will be those that dig deep to exploit every aspect of this new technology. The products we’ve seen so far have barely scratched the surface of what’s possible. There’s still no Uber of AI. But just as it took some time to mine the possibilities of the iPhone, the opportunity is there for those poised to seize it. “If you just hit pause on everything, we probably have five to 10 years worth of capabilities we could turn into new products,” says Josh Woodward, the head of Google Labs—a unit that cooks up AI products. In late 2023, his team produced Notebook LM, a writer’s support tool that’s way more than a wrapper and has won a rabid following of late. (Though too much of the attention has focused on a feature that transforms all your notes into a gee-whizzy conversation by two robot podcast hosts, a stunt that unintentionally underlines the vapidity of most podcasts.)
There are areas where generative AI has already made a very big impact. Coding stands at the top of the heap—companies now commonly boast that robots are doing 30 percent or more of their in-house engineering work. In fields ranging from medicine to grant-writing, AI has made a difference. The AI revolution is here, it’s just not evenly distributed. But for too many of us, taking advantage of the models involves crawling up a learning curve.
That’s going to change dramatically as AI agents perform all sorts of tasks, not the least of which is helping us tap the capabilities of AI without having to master prompt-whispering. (Though developers will have to negotiate the hard reality that granting agency to software robots is risky, particularly when AI is far from perfect.) Clay Bavor, cofounder of Sierra, which builds customer service agents for corporations, says that the creation of the most recent generation of LLMs proved to be an inflection point in the eternal quest for robots to act more like agents. “We crossed a critical threshold,” he says. Now he reports that Sierra’s agents can not only take a complaint about a product but order and ship out a replacement—and sometimes devise novel ways to solve problems that go way beyond their training.
When we look back on this year, the story probably won’t be about a single hot app but the sheer number of new tools that, in the aggregate, make a big difference. “It’s like asking, ‘What products are going to be invented with electricity?’” says Jang. “Will there be one killer app? Actually there will be a whole economy.”
So watch for a flood of new app announcements this year. And don’t write off the Googles, OpenAIs, and Anthropics as mere commodity providers. All of them are hell-bent on producing systems that make our current ones look as dumb as rocks—thus raising the bar for the next wave of app developers. I won’t dare make a prediction of what 2026 will look like.
Time Travel
I wrote about Sierra’s plan to put AI to use in customer service almost exactly a year ago, talking to its other cofounder, Bret Taylor.
Every time a new form of automation is introduced to shift the burden from humans to machines, companies must take care to soften the blow for customers. I am creaky enough to remember the advent of ATMs in the early 1970s. I was a grad student in State College, Pennsylvania. The entire region was flooded with ads—on billboards, in the newspaper, on the radio station—about welcoming “Rosie,” the name given to the machines being installed in the lobby of the biggest bank. (Even then, anthropomorphism was deemed necessary to soften the blow.) People eventually came to appreciate the advantages, like 24-hour banking and no lines. But it took years to trust those machines enough to deposit your check into one.
Taylor and Bavor believe that the transformative magic of AI is so good that we don’t need any softening. We’ve already been stuck with nightmare systems like phone support and websites that offer multiple-choice options that don’t address our concerns. Now we have an alternative that’s miles better. “If you survey 100 people and ask, ‘Do you like chatting with a chatbot?,’ probably zero would say yes,” Taylor says. “But ask the same 100 people, ‘Do you like ChatGPT?’ and 100 out of 100 will say yes.” That’s why Sierra thinks it can provide the best of both worlds: effective interactions that customers love, with the benefits of a no-downtime robot that’s not on the health plan.
Ask Me One Thing
Agoston asks, “Has your Roku been updated yet?”
Thanks for remembering my Roku issue, Agoston. To catch up the rest of you, just about a year ago I wrote a column about how some streaming services like Netflix consistently crashed on my smart TV with Roku. When I contacted the company, I discovered this was a known issue that Roku was taking its sweet time to fix. But their rep assured me that a fix was in the works, and one day soon an update would automatically install itself and make things right.
A few months later, what appeared to be an update process started on my screen, and I thought, finally I can watch more than two hours of Netflix or Hulu before the image freezes and I have to unplug the television set and reboot. For a while after that, I thought all was well. Maybe I just wasn’t watching much television. At some point the freeze came back—mostly on Netflix and sometimes on Amazon Prime or other services. I do not recommend smart televisions powered by Roku.
Submit your questions in the comments below, or send an email to mail@wired.com. Write ASK LEVY in the subject line.
End Times Chronicle
Vacation in beautiful Gaza, the new Riviera!
Last but Not Least
Bill Gates told me that Steve Jobs had a better batch of LSD than he did.
It’s not a crime to introduce you to the inexperienced youth squad that Elon Musk has unleashed on government IT services.
One of Elon’s protégés is only 25 and has direct access to the US payment system.
This Elon apparatchik is 19, his nickname is “Big Balls,” and he owns the domain Tesla.Sexy.LLC. Where have you gone, John Foster Dulles?