Agentic Commerce: The Four Levels of AI Shopping

AI shopping agent illustrating agentic commerce levels

Agentic commerce means AI agents that find, compare and buy goods or services for a consumer. It runs from a simple product recommendation to an agent that books a whole trip, depending on how much buying power the user hands over.

One useful way to see it is as four levels. Most of what exists today sits at the bottom two.

Shopping by remote has a long history. Merchants traded over long distances by letter and credit around 1900 BCE, Venetians sold books across Europe by catalog in the 16th century, and online payments, mobile wallets and crypto have arrived over the last thirty years. Agents are the next jump.

The difference from today’s recommendation engines is the willingness to act. Agents weigh price, specification and reputation, and with enough authorization they can negotiate and complete the purchase.

Level 1: discovery

The agent researches, compares products across retailers and suggests where to buy, using what it knows about you. It doesn’t place orders or negotiate.

That already beats a store’s own recommendations, which are limited to its catalog and your history there. ChatGPT, Gemini and Amazon’s Alexa for Shopping can do this today, though quality varies.

It sounds modest, yet it changes how people decide. Shoppers can weigh more options with far less effort, and although the agent doesn’t handle the transaction, it influences the decision.

Level 2: assisted purchasing

Here the agent helps complete the order: checking availability, entering delivery details and arranging payment. You must still confirm each purchase.

A retailer could receive and fill an order without the customer ever visiting its site. OpenAI launched in-chat purchasing through Instant Checkout in September 2025, then dropped it in March 2026 and sent shoppers to merchants’ own checkouts. Google’s Universal Cart, launched in the US in October 2026, collects items from several retailers and pays through Google Pay. Perplexity offers Instant Buy, and OpenAI has added ChatGPT Wallet for eligible US subscribers, with approval still required each time.

Level 3: conditional autonomy

At this level you set the conditions and the agent buys on its own once they’re met. Say you want navy Clarks desert boots in UK size 12 for no more than £65 including delivery. The agent watches prices and codes for days or weeks, then buys when an offer fits.

Trust becomes the problem. Retailers need proof an agent was authorized, and users need assurance it won’t overspend. Google’s Agent Payments Protocol (AP2) tackles this with cryptographically signed mandates that record spending limits and product requirements. Amazon’s Auto-Buy in Alexa for Shopping, which purchases an item at a target price, is an early, narrow version.

Level 4: orchestration

The agent coordinates several linked purchases toward one goal. You might ask for three nights in New York over New Year, leaving December 30, with flights, lodging and airport transport inside £1,500 per person.

The purchases depend on each other. Hotel dates must match flights, transfers must fit arrival times, and the total must stay under budget. If a flight disappears, the agent has to rethink several choices. This level is still largely aspirational.

Most of the progress so far has been gradual. Some purchases are easy to automate, while others carry financial, practical or personal considerations that keep a person involved.

What’s holding adoption back

A Boston Consulting Group study from October 2026 found only 4% of consumers would let an AI agent complete a purchase autonomously, even as more of them use AI for research.

Open questions include who is responsible when an agent buys the wrong item and how refunds and disputes work. Agents also need personal preferences to do well, and some people won’t want to share them. Retailers need systems that can talk to agents, and the infrastructure, including offerings from Stripe, remains fragmented.

Plenty of people also enjoy shopping, especially for expensive or personal items. Adoption probably won’t move evenly: someone may let an agent reorder household basics but still approve a new car or holiday personally.

What it means for retailers

If agents do the browsing, businesses compete for an agent’s recommendation instead of a visitor’s attention. Companies that rely on advertising or repeat site visits could suffer most, and retailers risk becoming plain suppliers if an assistant owns the customer relationship.

Traditional retail won’t vanish. Brand reputation, exclusive products, service and reliable fulfilment may matter more when agents compare offers instantly.

Source: jkfsom

Written by

Saksham Bhargava

Saksham Bhargava runs Alteroid and has been writing about consumer technology for over five years, covering iPhone, Android, Windows, AI tools and gaming. With an MCA and six years of hands-on web development experience, he tests the features he writes about on his own devices and focuses on clear, practical guides that solve real problems. Away from the screen, he enjoys travelling to places most tourists skip.