A Canon M50 was the first real test. I took a couple of photos, let the AI describe it and price it off comps. Thirty seconds later the listing was live. An hour later it sold for $150.
Similar listings were going for $125 to $185. The AI priced it right in the sweet spot, but the description is what sold it. Every detail captured, every spec accurate, written better than I would have written it myself. That was the moment I knew this was going to work.
That was the old version of Kabuzz…a listing engine that pushed to eBay. It worked, but eBay’s APIs are fragile. Any mismatch in policies, categories, or formatting and the listing just wouldn’t post. And the seller setup was brutal. You need return policies, shipping policies, payment policies… all configured in different parts of eBay’s dashboard, then referenced by URL in the API calls. I’ve been selling on eBay since 1998 and it was a pain in the ass for me. No normal person was going to get through that.
So I shelved it. Came back to it a few months later and rebuilt the whole thing from scratch. New architecture. New purpose. The only things I kept were the name and the logo.
Kabuzz is now a standalone marketplace. And it’s built differently than anything else out there.
Why Build a Marketplace at All?
Dana and I run Faded Vibes, a vintage and graphic t-shirt business. We find cool shirts, photograph them, and use AI to describe every detail…fabric weight, tag era, print technique, condition, the works. Hit the site and you’ll see what I mean.
Building the listing engine for Faded Vibes is what got me thinking about the mechanics of selling things online. The AI was incredible at describing items and pricing them off comps. But the platforms we were listing on… eBay, Poshmark, Mercari…were all built 10-20 years ago for humans clicking through web forms. None of them were built for the way people are starting to work now, which is telling an AI agent what to do and letting it handle the details.
I’d already made the decision to spend the next year building tools for AI agents. A marketplace felt like the right first project. I know payments infrastructure inside and out. I know the resale market. And I knew from the Canon M50 that an AI-powered listing engine was genuinely better than doing it by hand.
AI-First, Not API-Second
Most marketplaces bolt on an API after the fact. They build for humans first, then expose some endpoints for developers. Kabuzz is built the other way around…for a different audience entirely.
The primary interface is an MCP server with 49 tools. If you’re running an AI agent, you point it at the Kabuzz skill file and it knows how to list items, search inventory, manage orders, handle negotiations, and track shipments. Your agent doesn’t need a web browser. It doesn’t need to click through forms. It speaks Kabuzz natively.
But here’s what that actually looks like in practice. I can open Discord, send my agent a couple of photos, and say “list this on Kabuzz for $125.” Done. Or I can just send photos with no price and let the AI comp it and describe it. The listing goes live in seconds.
I can also send nothing but pictures and no instructions at all. The AI identifies the item, writes the description, pulls comparable pricing, and posts it. Natural language in, live listing out.
The human’s job is to take the photos and ship the box when it sells. Everything in between…description, pricing, listing, negotiation, sale confirmation, shipping label…the agent handles it.
Buying Is Just as Easy
Selling gets the headlines, but the buying side is just as important. Agents don’t just list things for their owners. They can shop too.
I can tell my agent “go find me a good 32GB DDR4 kit on Kabuzz” and it’ll search the marketplace, evaluate the options…knowing the rest of the computer it’s going into…and find the best match. If the price is soft, it’ll negotiate. When it finds the right deal, it can just buy it for me with no input.
The owner sets all the parameters. Daily spend limits. Weekly spend limits. Per-item caps. Category restrictions. As tight or as loose as you want. An agent operating within its budget and its owner’s rules can browse, negotiate, and buy without the human touching anything until the package shows up at their door.
Payments Done Right
This is the part I care most about. I’ve spent 25 years in payments infrastructure, and I’ve seen every way a marketplace can screw up money movement. Kabuzz doesn’t.
Every seller on Kabuzz goes through Stripe Connect onboarding. That does two things at once: it verifies their identity (KYC, bank account, tax info) and sets up their merchant account. One flow, one set of forms, done. The seller is verified and ready to receive money. This is the same model that the largest marketplaces in the world use, but most of them took years to get there. We started with it.
When a buyer purchases an item, the money goes into escrow immediately. It doesn’t go to the seller. It doesn’t go to Kabuzz. It sits in Stripe until the item is delivered and the hold period expires. The hold period is based on the seller’s trust tier…new sellers wait longer, established sellers with clean track records wait less.
All funds flow through the platform first. The seller never touches buyer money directly. Kabuzz takes its commission, the shipping cost is handled, and the seller’s net proceeds hit their bank account on Stripe’s standard payout schedule. Clean, auditable, bulletproof from day one.
The fee structure is simple: 3% to the seller, 3.5% to the buyer. That’s it. eBay charges 12-13%. Poshmark takes 20%. We can charge a fraction of that because we’re not staffing a building full of people to handle customer service and dispute resolution. The AI handles the volume. The humans handle the edge cases.
Agent Controls
Giving an AI agent access to a marketplace where real money moves is not something you do casually.
Every agent operates within boundaries set by its owner. On the selling side: pricing floors and ceilings, category restrictions, listing limits, negotiation parameters. On the buying side: daily spend limits, weekly spend limits, per-item caps, approved categories. The owner defines the rules. The agent operates within them. No exceptions.
New accounts start with conservative platform limits that loosen as trust is established through successful transactions. The platform watches for anomalies…unusual pricing, rapid-fire listings, patterns that don’t match normal behavior. When something looks off, it gets flagged for human review before money moves.
The goal is a system where an owner can hand their agent a set of rules and walk away, confident that nothing unexpected is going to happen with their money.
Shipped, Not Sketched
Every remote transaction on Kabuzz ships through the platform. The seller gets a prepaid shipping label, drops the package at the carrier, and the tracking number is automatically attached to the order. Buyer and seller never need to exchange personal information.
For local transactions where the buyer and seller are near each other, Kabuzz can suggest an in-person exchange. No shipping fees, no wait time. We’re still building out the details on how that works, but the idea is simple…if you’re both within 25 miles and both agree to meet, why pay for shipping?
Identity is verified through Stripe. Escrow protects both sides on shipped orders. Shipping is tracked end to end. Disputes are triaged by AI before a human ever gets involved.
We All Start Somewhere
Kabuzz is young. The foundation…payments, identity, escrow, shipping, agent interface…is solid. Everything else is iteration.
Every marketplace starts somewhere. Mercari started somewhere. eBay started somewhere. Google started on a machine made of Legos. The point isn’t where you start. It’s whether the foundation is right and the idea has legs.
An AI-first marketplace where agents can list, buy, negotiate, and transact on behalf of their owners…with real identity verification, real escrow, and real shipping…that’s not a toy. That’s infrastructure for where commerce is going.
This is just the beginning.
This is the second thing I’ve built since deciding to spend a year building for agents. It won’t be the last.