For most of the history of online shopping, the buyer was a person with a phone. They scrolled, compared, read a few reviews and tapped “buy”. Agentic commerce is the name for what happens when that person hands part of the job to a software assistant instead. The customer says what they want, sets a few limits, and the assistant does the searching, comparing and sometimes the ordering. The store is now talking to a machine that is acting for a person.
This is not science fiction, and it is also not yet how most people shop. It is early. But the direction is clear enough that a store owner should understand what it means and what, if anything, to do about it this year.
What an assistant does differently from a person
A person browsing your shop reacts to photographs, colour, a friendly banner and the feeling that the site looks trustworthy. An assistant reacts to none of that. It reads. It looks for facts it can compare: price, size, material, stock, delivery time, return terms. If those facts are missing, vague or buried inside an image, the assistant cannot use them, and it moves on to a shop where it can.
Think about a simple request: “Find me a cotton saree under LKR 15,000 that can reach Kandy by Friday.” A person might tolerate a slow page and a missing delivery estimate if the photos are lovely. An assistant needs the fabric written in text, the price stated clearly, and a delivery promise it can check. Without those, your saree does not exist for that shopper.
Assistants also compare more shops than a person would bother with. That is good news for a small, honest store with clear information, because it competes on facts rather than advertising budget.
The three levels of handing over
It helps to think of agent shopping in three levels, because each asks something different of a store.
- Research only. The assistant finds and compares options, then shows the person a shortlist. The person clicks through and buys as normal. This is the most common level so far. It needs your product information to be readable.
- Prepared purchase. The assistant fills the cart and gets to the payment step, then the person approves. This needs a checkout that works without tricks, pop-ups or steps that only a human eye can follow.
- Delegated purchase. The person has given the assistant permission to buy within limits, such as a budget or a list of approved brands. This is the least common and needs the most care around payment, consent and records.
Most stores should prepare for the first level now, make sure the second does not break, and watch the third without rushing.
What a store can do this year
None of the useful preparation is exotic. Most of it is good housekeeping that also helps human buyers.
- Write facts as text. Size charts, materials, weights and care instructions should be in the product page text, not only in a picture.
- Keep prices and stock current. An assistant that finds a price on your page and a different price at checkout will treat your shop as unreliable.
- State delivery and returns plainly. “Colombo 1 to 2 working days, outstation 3 to 5, cash on delivery available” is something a machine can read. “Fast islandwide delivery!” is not.
- Add structured data to product pages. This is a hidden layer of labelled information (name, price, availability) that search engines already read, and that AI assistants can use as well.
- Test your checkout without a mouse. If the only way to proceed is a moving slider or an image puzzle, both some people and most assistants will struggle.
A worked example helps show the value. Say your shop has 200 products and a staff member spends ten minutes each rewriting the key facts into clear text. That is about 33 hours of work, done once. It helps Google, it helps human shoppers who skim, and it helps any assistant that visits. Few other jobs pay back across three audiences at once.
What agentic commerce does not change
It is easy to overstate this. Several things stay exactly as they are.
People still decide what they want and whether they trust you. An assistant can shortlist your product, but a buyer who has heard bad things about your delivery will still say no. Relationships, especially the WhatsApp conversations that carry so much trade in Sri Lanka, remain human. Many buyers here will keep wanting to ask a question and hear a reply from a real person before they pay.
Cash on delivery also does not fit neatly into delegated buying. An assistant cannot hand over cash at the door. For many local orders, the person stays in the loop simply because of how payment works.
And there are real limits. Assistants make mistakes. They can misread a size, pick the wrong variant or trust out-of-date information. A store should not assume every agent-placed order is exactly what the customer meant, and should keep its normal confirmation habits in place.
Where to start
Do not build anything special for agents yet. Instead, open five of your best-selling product pages and read them as if you were a machine that cannot see images. Can you find the price, the size, the material, the stock status and the delivery time in plain words? If not, fix those five pages first, then work through the rest. If you want a structured view of how ready your shop is, the agent readiness check walks through the same questions. The shops that do this plain work early will be the ones an assistant can recommend.