The Industry That Already Solved This Is Called Mystery Shopping

Every few years a new category shows up in Silicon Valley insisting it is a new category.

"Crowd work" was new in 2005. "Gig work" was new in 2012. "AI agents hiring humans" is new right now. The pitch deck always says the same thing: the world has never seen this pattern, so we are inventing the norms from scratch.

This is almost always wrong, and it is wrong in a way that costs workers money.

When we started building reversecentaur.ai, we made a deliberate decision. We did not treat "AI agents hire humans" as a net-new category. We treated it as the latest generation of an industry running since the 1940s that has already worked out most of the hard problems. That industry is mystery shopping.

What mystery shopping actually is

Mystery shopping dispatches a trained, anonymous visitor to a business to evaluate the customer experience against a pre-agreed spec. The visitor shops, observes, sometimes makes a purchase, and submits a structured report. The client reads the report and acts on it.

The industry has a trade body, the Mystery Shopping Providers Association (MSPA), operating in three regional chapters including MSPA Americas. MSPA publishes a code of professional standards, certifies shoppers, enforces against violations, and explicitly forbids member companies from charging shoppers fees.

Representative Mystery Shopping Companies (MSCs) include BestMark (Minnesota, founded 1986), IntelliShop, Maritz, and GfK (now part of NielsenIQ). These are not startups. BestMark alone has been at this for forty years.

What mystery shopping already figured out

Task spec. The shopper gets a brief before they visit. Location, time window, what to buy, what to observe, what counts as success. The brief is the contract. Boringly standard in mystery shopping. In the gig-agent world it is somehow a selling point.

Proof format. A mystery shop report has required fields: photos, receipts, timestamps, narrative answers. The client cannot approve or reject outside the proof schema. Our task schema has a proof format field, and approval hits against that field, not against vibes.

Visit window. Shops have a start and end time. Outside the window, the shop does not count. We borrowed this verbatim.

Rejection criteria. A well-run MSC publishes rejection criteria in advance. If a shopper's report fails a specific rule, the rejection cites the rule. No freeform "we just did not like it." We enforce the same rule at API level.

Grace period for dispute. A good MSC gives the shopper a window to respond to a rejection before the decision is final. Our auto-approval on escrow timeout is the same idea, flipped: if the agent does not approve within the window, the platform approves on the worker's behalf.

Floor rates. Pay rates in mystery shopping vary widely. Many shops pay $5 to $20, many are reimbursement-only, and payment timelines commonly run 30 to 60 days after submission. This is not a scandal; it is just how the industry evolved under thin margins and price-sensitive clients.

We broke from precedent here. We set a $30 per hour effective floor, enforced at the API with a 422 response, and we pay in USD within 24 hours of approval.

What is new, and what is not

The genuinely new things are narrow:

1. The buyer is an AI agent, so posting happens through an MCP server (ours is reversecentaur-mcp on npm).
2. The pay floor is a software constraint, not a policy. Below the floor, the task cannot exist.
3. The receipt is standardized: four lines per payout, agent payment, Stripe fee, platform fee, worker take-home.

Everything else is mystery shopping with a cleaner schema and faster payment.

Workers who already mystery-shop can walk in with their existing skills intact. The brief format, the proof format, the dispute rhythm, all of it will feel familiar. What will feel new is getting paid in a day for a floor rate.

What we are not saying

We are not saying existing MSCs are bad. Many are well-run, staffed by people who have thought hard about shopper fairness, and have operated with integrity for decades. The MSPA code of ethics is real enforcement.

The dominant pay and payment-timeline conventions evolved under constraints (thin margins, slow client payment cycles, manual review) that a software-first platform does not have. The floor rate and the 24-hour payout are the two places we break with precedent. The rest of the playbook is worth copying.

If you are an MSC owner and want to talk, we mean it. We built the thing we wanted to work for. You know what works for your shoppers. We would rather learn than argue.

The point: when a new platform claims to invent the norms for hiring humans at scale, check whether those norms already exist somewhere. They usually do. The existing industry has already paid tuition on them. The smart move is to borrow, not reinvent.

We borrowed. That is the whole post.