Its 2025, and companies still dont know what AI is good for. Thats the impression I got from this years CES, which featured AI-powered kitchen appliances, baby cribs, and other products that really werent calling for AI.See: Spicerr, an intelligent touchscreen-equipped spice dispenser that learns your taste as you cook to recommend unique recipes.Spicerrs utility is a little questionable to begin with. Spicerr doesnt grind, and it takes $15-$20 proprietary capsules that cant be refilled. Setting all that aside, were people really itching for a meal-suggesting salt and pepper shaker to begin with?Elsewhere at the show, there was Dreos ChefMaker 2, an AI-powered air fryer. Yep, you read that correctly AI-powered air fryer. The concept isnt as outlandish as Spicerr, mind you. ChefMaker 2 can extract recipes from cookbooks via a page-scanning feature, and even handle the tricky math of calculating cooking times and temps. But is cookbook scanning really a feature the air-fryer-buying public demanded? Speaking as a member myself, I cant say its ever occurred to me and that appears to be true of most folks.Image Credits:DreoRemarkably, CES had even weirder AI products in store.Razers Project Ava, inexplicably named after the killer robot in the 2014 movie Ex Machina, is an AI gaming copilot, as the company describes it. Ava basically plays games for you without actually playing games for you. With permission, Ava captures stills of your computer screen, then gives pointers (e.g. Dodge when the blade spins). As The Verges Sean Hollister writes, Ava is controversial in that it was evidently trained on gaming guides, yet doesnt credit the authors. Its also distracting. At least in its current form, Ava is on a several-second delay, and it interrupts the games audio to give instructions.I must ask once again: Who was clamoring for this, exactly? Whos going to use it on a regular basis, much less pay for it? So far as I can tell, the out-there AI products at CES are a symptom of the industrys unrestrained hype. AI companies raised $97 billion last year in the U.S. alone, enough to buy 42 Spheres. Vendors are throwing AI spaghetti at the wall to see what sticks, because theres little downside to doing so and massive potential upside.In many cases, theyre also running up against the limitations of AI as we know them. Figuring out which use cases of AI are technically feasible has been a formidable challenge for the industry. Often, its led to over-promising under-delivering. ChatGPT still gets things wrong. Image generators are historically inaccurate. And characters in AI videos blend into each others bodies.So were stuck with IRL AI slop: air fryers, spice dispensers, and AI gaming copilots. Theyre not what most of us want, but theyre whats possible today with relatively low R&D lift. Heres to a better next year.TechCrunch has an AI-focused newsletter!Sign up hereto get it in your inbox every Wednesday.