Tanatasa Guides

What Is Agent-First Commerce? Why Not Just Let Agents Use Normal Ecommerce?

Follow one Tanatasa toothpaste-and-detergent order through a shopper, a browser agent and direct shopping tools. See what agent-first commerce adds to an ordinary online store.

What Is Agent-First Commerce? Why Not Just Let Agents Use Normal Ecommerce?

If an AI agent can open a website, select a product and click Add to cart, why does a store need to do anything differently?

It is a reasonable question. Agents can use ordinary ecommerce, and browser shopping remains useful. The case for agent-first commerce is about making a delegated purchase easier to complete, verify and hand back to the buyer.

Give Muse a concrete first task: Find fragrance-free laundry detergent and my usual toothpaste, delivered in Singapore, for less than S$40. Ask me before paying.

The agent needs to do more than navigate a page. It needs to identify the exact products, check quantities and stock, calculate the delivered total, prepare the right basket and distinguish a payment link from a completed order.

What does agent-first commerce mean?

Agent-first commerce is a design approach that treats a shopper's AI agent as an intended user of the store. Product information, shopping actions and transaction results are designed to be understandable and usable by software, alongside the experience for people.

We use the term here as a practical design description. It is not a certification or a guarantee of compatibility with every assistant.

For a shopper, it means you can describe what you want and have an agent help work through the purchase. For a store, it means providing clear product facts, dependable ways to change a basket, current totals and an appropriate handoff for approval and payment.

The website still matters. You should be able to inspect the products, change your mind and finish the purchase yourself.

A concrete example: the same Tanatasa order, three routes

Suppose you need two tubes of your usual toothpaste and one fragrance-free liquid detergent, delivered in Singapore for under S$40. Here is how the same order works through different routes.

Ordinary ecommerce, with you doing the shopping: open Tanatasa's shop, search for the toothpaste, choose the exact pack, add it to the cart, find the detergent and review the basket before checkout. You interpret product pages and operate the controls. This is the familiar product-page, cart and checkout experience of an online store; Shopify's Online Store guide describes that storefront model.

Ordinary ecommerce, operated by a browser agent: assign the list to Grok Bot or Perplexity Computer. The agent can work through the website using its browser tools: search, inspect a product page, choose a variant and add to cart. It still has to identify the controls and read the result of each interaction. These are examples of the task you would delegate, with the agent's browser as its way into the store. Grok Bot and Perplexity Computer explain their browser and computer workflows.

Agent-first commerce at Tanatasa: give Muse the same list and shopping quick start. Shopping with Muse at Tanatasa has been tested. Product search and product-detail tools return the exact variant, price, stock and pack contents. Muse creates a guest cart with the selected variant IDs and quantities, obtains a delivered quote and returns the shared cart link. You can open that same cart on the website. After your confirmation, checkout prepares the hosted Stripe payment page. Instinct, Dots, Claude Duet, Grok Bot, ChatGPT, Claude and Perplexity Computer can follow the same brief using Tanatasa’s merchant shopping interfaces and the tools available in the host.

The difference becomes visible with a twin pack. If one selected pack already contains two tubes, the agent should add one pack, rather than two. Structured pack details and a reviewable cart make that choice explicit. If the quote exceeds S$40, it should show the problem and ask what to change before preparing payment.

Tanatasa is both a normal online shop for people and an agent-first commerce example. The shopping interfaces add a direct route for software while retaining the website for browsing, basket review and payment.

Can agents already use ordinary ecommerce?

Yes. Browser and computer-use capabilities let agents read pages and interact with controls. That is valuable because it can work with stores that have not built dedicated agent tools. Anthropic's computer-use documentation describes both desktop interaction and browser-oriented tools, as well as limitations including latency and errors.

At Tanatasa, the shopping instructions explicitly provide a browser path when an assistant cannot use the shopping interfaces. It can browse product pages, use the ordinary cart and checkout, or return links for you to open.

For an occasional, straightforward purchase on a clear website, that may be enough.

Why build a more direct route?

Imagine an agent selecting a bottle size from a dropdown, dismissing a promotion, checking whether a subscription option is selected and navigating to delivery information. Each interaction requires it to interpret the current page and verify what changed.

A direct shopping tool can instead accept an exact product variant and quantity, then return the resulting basket. That can reduce the navigation and interpretation required. The benefit depends on the implementation; this article does not claim a measured speed or cost improvement.

Browser interaction is flexible. Structured operations make particular shopping steps explicit. A store can support both.

1. Product facts need to be unambiguous

A product title might mention a brand, scent, bottle size and promotional bundle in one sentence. A photograph might show packaging from another size. A low headline price might belong to the smallest option.

An agent needs to distinguish those details before comparing offers. Useful structured information includes the exact variant, pack quantity, price, currency, availability and relevant delivery or return information.

Shopify's catalogue documentation describes supplying titles, options, images, prices, availability and other attributes in a format agents can parse. This illustrates how product discovery can be designed for software as well as human browsing.

Structured data still needs to be correct. If the catalogue labels a scented detergent as fragrance-free, an API will pass along the wrong fact efficiently. Good data is part of the work.

2. Your brief needs to survive the shopping process

The instruction under S$40 delivered includes more than a product-price filter. The agent needs to account for quantity, applicable delivery charges and the final basket total.

Likewise, ask before substituting is a boundary on the task. Finding a similar toothpaste does not establish that you agreed to buy it.

A store cannot control every decision an external agent makes. It can provide enough detail for the agent to compare correctly, and require current information when preparing checkout. The agent and shopper still need to check the proposal against the original brief.

3. A basket needs a reliable current state

Suppose your agent prepares a basket while you open the same cart and change the detergent quantity. Which version should it use for checkout?

Tanatasa's shopping interfaces use versioned carts. An update based on an outdated version is rejected and returns the current cart, so the agent can read it again. Checkout also checks the expected total. These are concrete ways to detect a changed basket rather than continue from an old assumption. See the Tanatasa shopping guide.

The same cart can be opened on the website. That lets you inspect and edit what the agent prepared without starting a separate shopping session.

4. Checkout should make the approval boundary clear

Preparing a basket, approving a purchase, creating a payment link and confirming payment are separate steps.

Tanatasa’s agent checkout verifies current prices and stock, calculates the delivered SGD total and prepares hosted Stripe payment after buyer confirmation. A payment link is not a charge; the order status must confirm payment. Stock is not reserved merely because an item is in the cart. Read the checkout steps.

Agent-first commerce can also hand the buyer back to an ordinary checkout. Shopify's agent checkout documentation describes both agent-managed checkout sessions and handoffs to a merchant's prefilled checkout. The merchant remains responsible for the sale.

That handoff is part of the shopping experience. The agent can do useful preparation while the buyer reviews and completes the purchase.

5. Retries and interruptions need a defined result

What happens if an agent asks to prepare checkout and the connection times out? Repeating an action blindly can create confusion about whether it already succeeded.

Tanatasa supports unique request keys for checkout and payment preparation. Retrying the same request with the same key is designed to recover the existing operation rather than start a new one. Its order lookup provides a way to check what happened afterwards.

These details are mostly invisible when everything works. They become useful when a connection drops, an agent resumes later or the shopper asks whether an order actually went through.

Is this just adding an API?

An API is a way for software systems to communicate. It can help, but the quality of the commerce experience depends on what it exposes and how it behaves.

A complete shopping experience connects product search, cart building, checkout preparation and order confirmation. Clear instructions help an assistant use those steps together, while the buyer can review the basket and complete payment.

For shoppers, the useful test covers the whole task: correct product, correct quantity, current total, clear approval and a verifiable outcome. For merchants, the work includes product data, policies, access controls, error handling and continuity with the website.

When is ordinary ecommerce enough?

A clear storefront may be sufficient when the agent is researching a few products and returning links, or when you intend to complete checkout yourself. Browser shopping also provides a useful fallback when direct tools are unavailable.

The argument for dedicated interfaces becomes stronger when shoppers regularly delegate baskets, compare many variants, resume interrupted work or share cart editing with an agent. Stores should judge that need against their actual customers, traffic and implementation effort.

Agent-first design should make a purchase easier to understand and complete. It does not justify making the human storefront harder to use or claiming compatibility that has not been checked.

What should shoppers look for?

Ask the agent to show the exact variants, quantities, delivered total and delivery estimate. Check what you are approving and look for confirmed order status after payment.

At Tanatasa, start with the agentic shopping guide, with connection instructions and supported shopping actions for each assistant. Use the supported tools when available, and open the shared basket or product links whenever you want to review the details yourself.

Agents can use normal ecommerce. Agent-first commerce earns its value when it makes the delegated journey clearer, more dependable and easier for the buyer to stay involved in.

Sources and Tanatasa interfaces checked on 1 November 2026.