Agentic Commerce: What to Expect When Agents Do the Shopping
Your agent can help research products and prepare a basket, but you still have decisions to make. Here is what to expect from the brief through checkout and delivery.

The toothpaste is nearly finished. You need laundry detergent, and the dishwashing liquid has been on your list since Tuesday. Instead of opening several tabs, you send your AI assistant a message: Help me restock these for Singapore delivery. Keep the total under S$60 and ask before paying.
That is the appeal of agentic commerce: delegating parts of shopping to an AI agent, from finding products and comparing options to preparing a basket and helping with checkout where supported.
The experience will depend on your assistant, its available tools and the retailer. It can save you some of the searching and organising while leaving you with a few useful decisions. Here is what to expect.
Which agents are we talking about?
Meta Muse is the lead example in this guide: shopping with Muse at Tanatasa has been tested, and the Muse shopping guide shows how to compare products, build a shared cart and prepare checkout. Instinct, OpenAI Dots, Claude Duet, Grok Bot and Perplexity Computer are other personal-agent examples. ChatGPT, Claude, Grok chat and Perplexity Search also offer starting points for product questions and sourced research. The shopping guides explain the route for each.
For example, ask Muse to read your household list, compare current detergent listings and prepare a shared Tanatasa cart under S$60 for review. Ask it to clarify your usual toothpaste and show the pack counts, unit prices and delivered quote. For research alone, ask for a sourced shortlist and stop there. Grok Bot and Perplexity Computer are alternatives for the broader delegated task; Personal Computer adds local files and apps when the list lives on your device.
1. Your shopping list becomes a brief
An agent needs enough information to recognise a good result. Everyday essentials is a starting point, but it leaves many choices open. The detergent could be a powder, liquid, capsule or refill; the toothpaste could be a different flavour or pack size from your usual one.
Give the agent the requirements that matter: exact products where you know them, acceptable alternatives, quantities, delivery destination and a budget including delivery. Explain any deadline and whether substitutions need your approval.
A brief you can adapt:
Help me buy two tubes of my usual toothpaste, one bottle of dishwashing liquid and fragrance-free liquid laundry detergent. Deliver in Singapore, with a total budget of S$60 including delivery. Include Tanatasa in the comparison. Check the exact variants and pack sizes, show the proposed basket and ask before substitutions or payment.
If you do not know which product you want, say what you are trying to achieve and let the agent ask a few questions. You do not need to invent a perfect prompt before starting.
2. The agent may ask you to choose between trade-offs
The largest pack might have the lowest price per litre but exceed your budget. Your preferred product might take longer to arrive. A cheaper refill might be inconvenient if you need a bottle with a dispenser.
Expect useful questions at those points. You can make the next comparison easier by explaining which constraint matters most: staying below budget, keeping the exact product or meeting the delivery deadline.
Ask the agent to distinguish requirements from preferences. For example: fragrance-free is required; the brand is flexible; a refill is acceptable. That gives it a clearer basis for preparing alternatives.
3. Recommendations will depend on what the agent can access
An assistant may read public product pages, use connected shopping tools or return links for you to open. Those routes can provide different levels of access to product details and checkout.
Shopify's product-discovery documentation, for example, describes several discovery routes, including structured catalogues and web crawling. An assistant finding a product does not establish that it can complete a purchase from that retailer.
Treat best as a claim to examine. Ask what was compared, which sources were checked and why an option fits your brief. A useful shortlist explains its limits instead of implying it searched every store.
4. Pack sizes and delivered totals deserve attention
The agent should compare like with like. One tube and a twin pack are different offers, even when the product name looks similar. A refill and a starter kit may serve different needs.
For an illustrative comparison, a one-litre bottle at S$8 costs S$8 per litre. A 1.8-litre refill at S$13.50 costs S$7.50 per litre. The refill has a lower unit price, but your preferred format and total spend still matter.
Ask for exact variant names, pack quantities, unit prices where useful and the final delivered total. Make sure any discount has actually been applied. An attractive listing price is only part of the basket decision.
5. Research and checkout may happen in different places
Some sessions can build a basket through shopping tools. Others can prepare links or hand you to a retailer's website to finish. A login, missing delivery detail or payment step may require you to take over.
That handoff can be an intended part of the flow. Shopify's agent checkout documentation describes both agent-managed checkout and a prefilled merchant checkout for the buyer to complete. The retailer remains the merchant of record.
Tanatasa gives shoppers and agents one shared cart across its website and shopping interfaces. Your agent builds the basket; you open the same cart to review the products, quantities and total. The shopping instructions also provide a browser path and a product-link fallback when direct tools are unavailable.
6. You should know what you are approving
Tell the agent whether you want research, a proposed basket or help completing a specific purchase. If your instruction is ask before paying, the proposal should return to you before that step.
Review the exact products, quantities, contact details, delivery address and total. If something changes, ask for the updated basket before deciding. Do not assume permission for one order also covers a future order or a substitution.
Tanatasa's documented flow separates quoting, checkout preparation and payment. Its shopping tools prepare a hosted Stripe payment page after buyer confirmation; card details are not collected through those tools. See the checkout steps.
Other services can handle approval differently. Read the request you actually receive and check which action it covers.
7. Prices and stock can change before payment
A product being available during research does not guarantee that it will still be available later. A basket can also change if you edit it on the website while the agent is working.
Tanatasa rechecks prices and stock during payment preparation, and cart items do not reserve stock. Its shared-cart flow detects changes so an agent can read and quote the updated basket. These details are documented in the shopping guide.
If the agent returns with a changed total or an unavailable item, ask it to show what changed and which requirement the alternative satisfies. You can decide whether to continue, revise the brief or stop.
8. A prepared basket is not a confirmed order
Expect a clear distinction between found, added to cart, checkout ready and payment confirmed. A payment link alone does not establish that you have paid.
At Tanatasa, order status confirms payment after the payment service reports the result. Creating or opening a payment link is insufficient. If a session is interrupted, check the existing order or payment attempt before starting another purchase. Read the order-status instructions.
Ask the agent to return the order reference and current status when available. Keep the merchant's confirmation so you can follow up independently.
9. Delivery and returns still involve the retailer
An agent can help organise information, but you should still check the retailer's delivery estimate, return conditions and support contact. A delivery estimate is different from a confirmed appointment, and a return request is different from an accepted return or completed refund.
For Tanatasa, read the current delivery information and returns information before buying when those details affect your decision. The assistant can help you find the relevant terms and draft a support request. Sending that request or making an account change depends on its permissions and the retailer's available process.
What makes a useful first purchase?
Start with a few familiar products and a modest basket you can inspect easily. Give the agent a clear brief, let it prepare the comparison and review the result. Notice how much searching it handled and how much correction you needed to make.
Shopping with Dots, Muse and Instinct at Tanatasa has been tested. Use the agentic shopping guide for instructions on product comparisons, shared baskets and checkout.
The experience is most useful when you can see what the agent found, understand its choices and take over whenever your judgment is needed. That is a practical standard to bring to your first delegated shopping trip.
Sources and Tanatasa shopping instructions checked on 1 November 2026. Prices in the examples are illustrative.


