It’s impossible to predict all the ways in which AI will change our lives, but one sure bet is that virtual assistants will help us navigate this brave new world. Some will be built into our smartphones, but companies big and small also hope consumers will find space for an extra AI-powered gadget—and perhaps eventually ditch their phones altogether.

The category has not started off so well, as some of the most eagerly awaited AI-powered devices to land this year have done so with a resounding thud.

First, there was the Humane AI Pin, a $699-plus-$24-a-month lapel pin that projected images onto a user’s hand. The matchbook-size device was supposed to wean people off their smartphones, but was instead lambasted by reviewers for its cost and poor performance. Then came the Rabbit R1, a $199 handheld gadget that could answer questions and identify things its camera saw. Critics were kinder to the Rabbit, but still asked what its point was, since our phones can now do much the same.

So what is the ideal AI gadget? Is the optimal design something that unleashes an all-powerful AI genie? Or is it one of our existing accoutrements, made more useful with a dash of AI?

There’s a broad “scramble within the consumer electronics industry to find the best applications for new large language models [LLMs] within hardware,” says Jack Leathem, a research analyst at Canalys. Expect a lot of trial and error as companies compete to devise the winning recipe.

A gadget with ears and eyes

Some are betting the answer is to focus on specific tasks. Limitless and Tab AI, two venture-funded startups, are developing smart pendants that will listen to the wearer’s conversations and transcribe them into digital form.

Apple and Google are looking to turn existing wearables like earbuds and fitness bands into AI gadgets. In the case of Apple’s wireless AirPods, that could entail integrating tiny cameras into the earbuds so that the AI can respond to all manner of stimuli, Bloomberg reported.

One promising approach comes from Facebook parent Meta. Late last year, Meta launched the second generation of its collaboration with French-Italian eyewear giant EssilorLuxottica: the Ray-Ban Meta smart glasses.

The Wayfarer version looks virtually identical to the classic shades sported by Tom Cruise, Taylor Swift, and other pop culture icons. Tucked inside the frames, however, are a tiny camera, a wireless chip, and other components that allow users to broadcast live video streams and listen to music. In April, Meta added AI capabilities to the North American version—a small but crucial tweak that transformed the specs from a geeky novelty item into the foundation for a hands-free superpower.

With AI aboard, the glasses offer a glimpse of the game-changing potential in “multimodal” AI: Wondering what a good recipe might be for that fresh slab of salmon you found at the market? Just look at it and summon the AI’s expertise by uttering a few words—no need to pull out your phone, or type into Google. The experience can still be a bit clunky, according to reviews—zooming in on a specific object is not easy, and the AI itself is as error-prone as any other chatbot—but the potential is clear.

“We have the leading AI device on the market right now, and we are doubling down on finding strong product-market fit for wearable Meta AI, building a business around it, and expanding the audience,” Meta chief technology officer Andrew Bosworth wrote in an internal memo in June, according to the Verge.

How much would you pay?

Meta hasn’t disclosed the number of smart glasses it has sold. Forrester analyst Thomas Husson believes connected glasses are still a niche product and notes that gadgets are not Meta’s strong suit. “It’s a way for Facebook to showcase its technology and to appear innovative,” he says.

If Meta’s glasses catch on with the masses, however, Canalys’s Leathem says, the company could reap big benefits: “This integration would position Meta’s AI as the primary assistant, rather than a secondary one within another vendor’s device.”

Bobak Tavangar, CEO of Singapore-based Brilliant Labs, says AI gadgets are so personal that they must be “open and modifiable and inspectable.” That means open-source designs and a choice of LLMs. His company’s recently launched Frame smart glasses feature an AI assistant that draws its powers—including translation, search, and visual analysis of one’s surroundings—from AI models made by OpenAI, Stability AI, and Perplexity.

Brilliant Labs sold thousands of the $349 smart glasses in the first couple days, Tavangar says. He believes that companies with advertising-based businesses, like Meta, are ill-suited for the new generation of AI devices, since consumers will worry that personal data captured by the spectacles will be fed into a giant advertising engine.

Meta didn’t respond to Fortune’s question about using smart-glasses data for advertising purposes, though its privacy policy appears to allow it. Indeed, The Verge reported Friday that Meta’s big rival in online advertising, Google, has also been talking to EssilorLuxottica about putting its Gemini AI assistant into future shades. Multiple outlets have reported that Meta is preparing to buy a stake of roughly 5% in the eyewear house.

As AI gadgets evolve, business models could prove as important as hardware design: After all, if a gadget maker isn’t earning ad revenue, it will likely have to sell the product at a higher price or, as Brilliant Labs intends to do, charge a monthly fee for access to AI services.

AI can do a lot of nifty things, but it can’t eliminate economics—yet.