One of the promises of the next era of generative AI is that the technology will be agentic, or have the ability to perform tasks autonomously on behalf of us chaotic humans. That means AI agents will theoretically be able to “reason” about the next steps they should take, allowing them to execute multiple actions from a single query. The possibilities are endless, if you believe the hype—think maximum efficiency and productivity, plus a host of other buzz word-latent phrases that one might hear during a tech giant’s quarterly earnings call. All I want AI to do for me, however, is to shop.

I understand that some people find shopping to be a pleasurable act, but the options overwhelm me, whether I’m in an actual store or stuck in an endless scroll. In the lead up to the December holidays, the pressure mounts even more: How do you convey to someone exactly how much you’ve appreciated them this year—all the years—and capture that appreciation in something more thoughtful than a soy wax candle? I was ready to let AI take the wheel.

Over the last few weeks, I’ve offloaded my gift shopping to Perplexity AI, OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, and Amazon’s Rufus to find out whether I could finally delegate one of the tasks I despise the most to AI. I used the apps as pure utilities; for literal fulfillment, means to a commercial end. I set the generative AI bots loose and burned planet Earth in a quest to find the perfect baking equipment.

Spoiler: I quickly learned that none of these apps can autonomously shop, at least not yet. They are glorified search tools that have the ability to parse and summarize product descriptions, as well as compare different items. I still had to write and rewrite prompts for what kinds of gifts I was looking for, and for the most part, I also needed to enter in my payment information and go through the purchasing process on each individual retailer’s website.

I used the bots to shop for five people, ranging in age from 6 months to 49 years old. A close friend who is extremely into baking became one of my primary test cases. I also tried to drum up holiday gift ideas for my 16-year-old niece, who once complimented me in a text message by saying, “Don’t worry, you’re not mid.” (I have the screenshot saved.) Another person on my shopping list was an editor and musician friend with eclectic taste who’s celebrating a milestone birthday just after the new year.

One of the AI apps I tested has a feature specifically designed for shopping. Perplexity AI, a well-funded generative AI search startup that has been criticized for allegedly lifting content from news publishers, rolled out a new service last month called Buy with Pro, which comes with a $20 per month subscription to Perplexity Pro. Buy with Pro is described as a “first-of-its kind AI commerce experience” that promises to make shopping online “10x more easy and fun.” This immediately sounded like an AI hallucination to me, because online shopping isn’t fun in the first place. (In terms of disclosure, Buy with Pro and other AI gift guides are also technically competitors with WIRED, which earns affiliate revenue from our entirely human-produced, human-reviewed, and human-edited gift guides.)

When you run a shopping-related query on Perplexity Pro, the app shows that it’s “reading” sources like The New York Times, The Food Network, Reddit, and others. Several seconds later, it presents a cascade of products, along with prices and retailers. Some of the listings now have buy buttons, powered either by Shopify or by Perplexity’s own payment processing service. If you click on one and order something, the shipping is free. There’s also a visual search tool within the Perplexity app, which lets you take pictures of things and look for similar items for sale online. (Perplexity says it isn’t collecting affiliate revenue from sales made through its platform.)

I tried searching Perplexity Pro for the “ideal holiday gift for a close friend who likes to bake but already has everything.” The AI produced a list of items that mostly fell on what I would describe as the lowbrow-useless approval matrix, with occasional bits of brilliance. It included a $10 Tasty Tinies miniature baking set for children (not applicable), a $120 Bakken-Swiss stackable 8-piece bakeware set (possibly useful), and a $35 sweatshirt with the phrase “My Buns Are Gluten-Free” printed on it (just no). Tweaking the prompt to add words like “luxury” or “customized” did improve the outputs slightly, but not by much.

Using Perplexity’s shopping feature quickly started to feel not much different from browsing Amazon or Walmart’s websites—or maybe a product reviews site—just wrapped in a C-3PO, let-me-compute-for-you skin. Amazon’s Rufus AI also provides this kind of service right on Amazon’s website and in its mobile app, where there’s a bot that can answer questions, compare products, and help you mainline more junk from Amazon. When I asked the same baking gift prompt to Rufus, it immediately suggested that I purchase my friend a KitchenAid Stand Mixer that costs upwards of $300. (Rufus seemingly assumes you have Jeff Bezos bucks.)

I then turned to the three other AI chatbots, none of which have standalone ecommerce features. But one of the main selling points of tools like ChatGPT is that they are supposed to help people brainstorm and come up with ideas—exactly what I wanted.

When I asked the same question about what to get my friend who loves baking, ChatGPT’s responses were the most thoughtful and inventive. It provided 15 different gift ideas, categorized by themes like Baking-Theme Gifts, Luxe Home Items, and Personalized. A handmade ceramic mixing bowl? A premium tea or coffee sampler to go with baked goods? A recipe journal to document adventures in baking? This was the good stuff.

In its initial responses, ChatGPT didn’t provide any links to products. But it easily supplied them when I asked, and while I didn’t click on every single one, none appeared to be hallucinations. Claude, on the other hand, apologized and said that it “cannot actually link to websites or products directly.” Anthropic hasn’t released a web search feature for Claude yet, but the company says it’s working on it.

That technically made Claude the least useful chatbot I tested for shopping. But it also means that Anthropic has so far avoided wading into the ethically murky territory of allowing its AI chatbots to scrape human-written product reviews from the web. Instead, Claude bases its product comparisons on its existing data set. Perplexity, on the other hand, says that thanks to Buy with Pro, people “no longer have to scroll through countless product reviews.”

When I asked Perplexity what I should get for my editor/musician friend, it recommended a solar bike light set (I also noted he was a cyclist). It wasn’t a bad idea, but not exactly a milestone-birthday worthy gift. I kept tweaking my prompt. What about a personalized leather guitar strap? Down the rabbit hole I went.

Perplexity’s goal in hyping up its shopping features, I was beginning to understand, wasn’t just to help me brainstorm fresh ideas or come up with supremely thoughtful gifts. Perplexity is playing the long game, slowly siphoning our attention away from competing corners of the web, gaining a better understanding of how people like me are using its platform, and funneling that data into its ever-evolving AI models. Each time I needed to refine my searches because the initial results were often lacking, I remained in Perplexity’s app, which meant I was not on Amazon and not on Google (though I ended up on both of those sites eventually). Perplexity Pro is not a full-fledged ecommerce site, nor is it “agentic” in any real way yet, but I am one of millions of people supplying the information it needs to become those things.

When I turned to Google’s Gemini, I found the gifts it suggested for my 16-year-old niece weren’t bad, per se, just uncreative and, in one instance, confusing. It said I should buy her a “cat blanket for snuggling up with a good book,” but it wasn’t clear if the blanket was for her or her cat. A Kindle was a fine idea. But I’m terrified of what she would text me if I sent her the SAT prep book Gemini suggested (probably “thx,” and nothing else). The app’s ideas for my editor/musician friend were equally uninspiring, among them “Vinyl records,” and “High-quality headphones.”

I was using the year-old version of Gemini, but earlier this month, Google started rolling out a newer version, Gemini 2.0, to developers and limited testers. The new AI model will “think multiple steps ahead, and take action on your behalf,” the company says. For now, this means taking action on behalf of developers—executing the next step in their coding workflows—but I’m eagerly awaiting the day it can plow through my shopping list.

ChatGPT eventually led me to an online spice store where I bought a few specialty baking ingredients for my friend, who at this point, I had built up in my mind to be a finalist in The Great British Bake-Off. In the end, I chatted with the AI bots for so long that many of the gifts I picked won’t arrive until after Christmas. My niece will be getting cash in a card. My search for a friend’s milestone birthday gift was inconclusive. I decided to kick the task down the road until January, a month full of newness and agentic resolve.