Your Payment Gateway Just Hired You an AI Team: PayU's Agent HQ for Small Sellers

PayU's Agent HQ: an agent store for small sellers

In early September 2026, the fintech company PayU announced Agent HQ, described as an AI agent store built for Indian small and mid-sized merchants. The idea is simple: instead of buying one big AI product, a seller picks individual AI agents, each hired for a specific job in the business. One agent works on getting the store discovered by AI shopping assistants. Another handles payment-related everyday operations. A third powers the store's own conversational assistant. PayU's announcement frames the goal as two outcomes a merchant can pick: build what it calls an AI-ready growth engine, run the business more efficiently, or do both.

Why this launch matters to a small seller has less to do with the technology and more to do with who is offering it. PayU already processes payments for over 4.5 lakh merchants and sees hundreds of millions of shopper journeys across its platform. That means its agents start with a decade of commerce intelligence: what real orders, refunds, and settlements look like for businesses like yours. When an agent drafts a response to a chargeback or flags an unusual settlement, it is drawing on patterns learned from thousands of similar sellers, not starting from zero.

The five jobs on the agent team

PayU's announcement groups the work into a handful of jobs that map surprisingly well onto the chores a small seller already does. Read the list and notice how ordinary each one is:

First, discoverability. As more shoppers ask AI assistants where to buy something instead of scrolling through search results, stores need to show up inside those answers. One job on the agent team is making the business discoverable to AI shopping agents — keeping product information complete, current, and in a format assistants can read.

Second, growth. Agents that work on driving sales: following up on abandoned carts, nudging repeat buyers, surfacing offers to the right customers at the right time.

Third, payment operations. This is the unglamorous heart of the announcement: reconciliations, settlements, refunds. Every day, money lands in a seller's account from multiple payment methods, and someone has to check that the numbers match what was actually sold. Agents that watch these flows can flag a settlement that looks wrong before it becomes a week of confusion.

Fourth, chargebacks. When a customer disputes a payment, the seller has a short window to respond with evidence — order details, delivery proof, refund records. Miss the window and the money is gone. An agent that gathers the evidence and drafts the response turns a panic job into a routine one.

Fifth, the store's own conversational assistant: a chat experience the merchant controls, answering buyer questions and taking orders, built on the seller's own catalogue rather than a generic script.

None of these jobs requires the seller to write a line of code or train a model. That is the point of the store format: pick the job that hurts most, deploy that agent, ignore the rest until you are ready.

The practical use case: one week at a home-goods store

Take a small home-goods store selling kitchenware online — the kind of business doing a few dozen orders a day with two or three people handling everything. Here is what a first week with one agent from the team could look like.

Pick the chargeback-and-refund agent first, because it touches money directly. On Monday, connect it to the store's payment account and order records. Set one rule with it: draft every response, send nothing. You review and approve each draft yourself for the first two weeks.

On Tuesday, a customer claims a dinner set never arrived. The agent pulls the order, the courier tracking, and the delivery photo, and drafts the response with all three attached. What used to be forty-five minutes of digging through three different screens takes you five minutes of reading and one click to approve.

On Wednesday, the agent flags something you would not have caught: two settlements from the same day that do not match the day's sales, a small shortfall. You check with the payment provider and find a settlement split you did not know about. It is not fraud; it is a new payout schedule. But now you know, instead of discovering it at month-end.

By Friday, the routine is set. Refund requests get drafted the same day instead of piling up for the weekend. The agent keeps a log of everything it drafted and you approved, so there is a paper trail if a dispute escalates. You have not given it the power to move money — you kept that approval for yourself — and it has still taken the most annoying hour of each day off your plate.

That is the honest shape of this technology right now. It does not run your store. It does the digging, drafting, and flagging; you keep the decisions.

Why sellers are reaching for agents before hiring

There is a reason agent stores like this one are launching now: small business owners have quietly changed how they think about getting help. In September 2026, the accounting software company FreshBooks published a survey of 500 American solopreneurs and microbusiness owners, and the numbers are striking. When these owners hit a business task they could not complete on their own, 86 percent said they try AI before hiring someone to do it. Nine in ten said AI makes it easier for one person to start and run a business.

The most telling number is about what they actually want from the technology. Asked to choose, 64 percent picked the ability to appear bigger or more capable than their actual size, while 36 percent picked getting more done without adding time or cost. For a three-person store, that reads as: handle like a team of ten, stay a team of three.

That instinct has a catch, and it is worth naming before you deploy anything. An agent that drafts chargeback responses or reconciles settlements is only as good as the records it can see. If your order data lives in three places with three different spellings of the same product, the agent will multiply the confusion, not fix it. The sellers who get the most out of these tools are the ones who clean up one data source first — usually the order records — and then let the agent loose on it.

Three rules before you switch your first agent on

First, start with one pain, not five. Pick the job that costs you the most time or the most money when it goes wrong — refunds and chargebacks are the usual suspects for sellers. Run that one agent for a month before adding a second. An agent team of five that you never check is worse than one agent you supervise properly.

Second, keep the approval on money. Let the agent draft, flag, and prepare, but do not let it issue refunds, accept chargeback losses, or change payout settings on its own — not at first, and arguably not ever. The few minutes of review are the cheapest insurance in the business.

Third, treat onboarding like training a new employee. Give the agent clean records, clear rules for what it may and may not do, and a weekly ten-minute review of what it handled. The stores that treat deployment as a one-click install get one-click results. The ones that spend an hour setting it up properly get an actual team member.

Agent stores like PayU's will keep growing — more jobs, more specialised agents, more competitors launching their own versions. The sellers who benefit will not be the ones who deploy the most agents. They will be the ones who pick the right first job, keep the approvals, and check the work. That discipline has always separated good operations from chaos; AI agents just make it cheaper to practice.