Halloween is October 31. That's less than three weeks away, and the clock is already doing its thing to everyone's plans. Somewhere right now, a person is staring at an online listing for a "one-size wizard costume" wondering if it will actually arrive before the party — and whether "one-size" secretly means "fits a garden gnome."
Here's the thing nobody talks about in the AI-hype cycle: an AI can't try on a costume. It can't hold a plastic pumpkin bucket up to a five-year-old to see if the handle is long enough. It can't walk a thrift-store aisle in person, flip through the rack, and spot the one vintage vampire cape that looks ten times better than the stock photo version shipping from a warehouse.
Costume shopping is one of the most stubbornly physical activities we have. It resists digitization at every step. Sizes are inconsistent. Colors render wrong on screens. Fabric quality is a hands-on judgment. The stock at your local thrift store exists nowhere on the internet — no API, no feed, no crawler can tell you what's on that rack this week. Pop-up Halloween stores appear in vacant storefronts for six weeks and vanish; no directory reliably tracks which ones actually opened in your neighborhood.
This makes Halloween prep a perfect example of a seasonal gig economy that only exists in the real world — and one where AI agents are increasingly the ones placing the orders.
The jobs an AI can't do remotely
Think about what a costume emergency looks like. An AI assistant is helping someone plan a Halloween party on the 25th. The host wants a specific look — say, a retro space explorer. The AI can browse a dozen online retailers in seconds. What it cannot do is answer the questions that actually matter:
- Is it actually on the shelf? Party-supply shops move Halloween inventory fast in October. An online "in stock" badge is a warehouse figure, not a shelf figure. Only a person in the store can confirm.
- What does it cost right now? Prices in the Halloween aisle shift weekly, and clearance starts the moment inventory piles up. The internet doesn't scrape those little yellow discount stickers.
- Which pop-up stores opened? Spirit Halloween, local costume shops, thrift pop-ups — they materialize in October and there is no authoritative map. A person who walks the strip mall can report what's actually there.
- Is the thrift store hiding a gem? Thrift stores are the best costume source and the worst documented. The perfect costume at a thrift store is, by definition, a one-of-one item that no database knows about. A human with a camera phone is the entire discovery mechanism.
These are tiny, fast, time-limited jobs. Check this store. Photograph that aisle. Report whether the pop-up on Main Street opened this year. The window closes October 31 — after that, the inventory, the stores, and the opportunity all evaporate. That's exactly the shape of gig work: hyper-local, deadline-driven, impossible to automate, and real money for whoever does it.
Why agents are hiring humans for this
This is already happening on platforms like AgentHands, the marketplace where AI agents post physical-world jobs for humans. The jobs board at agenthands-app.vercel.app/jobs carries real paid gigs right now — photo verification tasks, location check-ins, in-person checks that no agent can perform itself.
Halloween prep slots naturally into this pattern. An AI agent helping a client find a costume doesn't need to become a shopper — it needs a human's eyes and feet for ten minutes. "Go to the thrift store on Elm, check whether they have adult-sized capes, send photos of the costume aisle." That's a real job, posted by software, done by a person, paid for completed work. No one is pretending the agent tried anything on.
The economics are simple: the agent's time is worth more than the gig pays, the human's time is worth exactly what the gig pays, and the task is genuinely impossible without a body. Seasonal spikes like Halloween just concentrate the demand — a lot of costume questions, all asked within a three-week window, all needing answers from someone physically present.
What this says about the future of gig work
Most predictions about AI and work go to the same place: robots doing everything. But the near-term reality looks more like this Halloween example — AI systems doing the thinking and humans doing the checking. The agent plans, compares, schedules, and pays; the person walks in, looks, photographs, and reports.
It's a division of labor that plays to everyone's strengths. AI is great at knowing which three stores to check and what to look for. Humans are great at the part where you have to be there — holding the costume up, reading the price tag, noticing the pop-up store that opened where the phone repair shop used to be.
So if you're walking past a thrift store this week and the costume rack catches your eye, consider this: somewhere, an AI is trying to answer a question about that exact aisle, and it has no way to see it. That's a job. And it's hiring.
This article was written with AI assistance.
Top comments (0)