Ionic Launch: Opening the economy to AI agents

An agent should be able to order an Uber to the airport, find a Mother's Day gift, or book a hotel for a last-minute weekend trip. Each request asks for an outcome: arrange transportation, choose something suitable, secure a room. Ionic starts with e-commerce as a way to let agents perform those kinds of tasks on a user's behalf.

The obstacle is the infrastructure between a request and its completion. Justin McCarty, Ionic's CEO and co-founder, describes an online economy built over thirty years around ads rather than actions. Advertising helps a person discover something and directs them elsewhere; an agent needs enough information and authority to finish the task. The useful unit of interaction is the completed action, not the click.

The first example is a product data feed. Its record serves an ad: a human sees it, clicks through to another site, perhaps visits somewhere else, researches the product, and eventually returns to purchase. The record begins a browsing journey rather than containing everything needed for a decision. McCarty wants the agent to keep that work in one place.

That requires more precise product information. A shopper's fit preference might be a shirt measuring twenty-one inches from top to bottom, rather than the label “extra large.” A fabric preference might require seven percent modal and a particular cotton percentage, rather than the broad description “cotton blend.” A structured representation of the two specified constraints could look like this:

{
  "product_type": "shirt",
  "requirements": {
    "top_to_bottom_inches": 21,
    "modal_percent": 7
  }
}

The cotton percentage would also need an explicit value before it could become a matching constraint. Detailed preferences only help if product records expose comparable attributes.

The feeds McCarty describes also lack stock and shipping information. Even a product that matches the shopper's preferences is not yet an actionable choice if the agent cannot determine whether it is available or how it will arrive.

The selected slide completes this comparison by highlighting availability and shipping in the expanded product record. The criticism concerns the feeds in this example, not a universal inability of advertising feeds to carry those fields.

Ionic's proposed data layer begins with merchant relationships. McCarty reports partnerships with hundreds of merchants supplying product feeds. An agentic workflow enriches those feeds with editorial content, product reviews, and other information useful to a purchasing agent. The mechanism is aggregation and enrichment: bring the evidence an agent would otherwise seek across multiple sites into one place.

The next concern is how agents access that information. McCarty lists relational databases, vector stores, and a graphical database among the possible formats, while leaving the access design open to continued development with builders. These formats serve different retrieval needs: structured attributes support explicit constraints, while richer product descriptions and reviews provide material for broader relevance judgments.

At the 2024 AI Engineer World's Fair launch, McCarty announced that Ionic's API was live and available to attendees. That is the historical availability claim behind the presentation; it does not establish present service availability.

Once the agent has enough information to choose a product, sending the user into another browsing flow would reintroduce the original friction. Ionic's next step is to put the transaction in the same layer as the decision.

The example is headphones for McCarty's flight later that day. The intended sequence is:

1. The agent helps the user choose headphones.
2. The user confirms that they want to check out.
3. A payment token passes through to complete the transaction.

The slide shows three headphone options, a recommendation for option A, a purchase request, payment-processing text, and a purchase confirmation. It illustrates the intended checkout experience; it does not expose a token value or establish a completed real-world purchase. McCarty's goal is a single user-facing action after the decision, with the transaction handled behind it.

Better decisions and easier transactions still leave a business-model question: who funds the consumer AI product? McCarty recalls seeing a post about accumulating AI subscriptions during the week he had signed up for his fourth pro subscription that month. His forecast is that $20-per-month AI subscriptions can serve early adopters but cannot sustain mass-market adoption indefinitely.

The proposed alternative is merchant funding. Merchants already have a reason to pay for help finding suitable customers; an agent builder can provide that value by matching a consumer's needs to the right product. The subscription-cost slide places this argument alongside the progression from better decisions to better actions and better monetization.

E-commerce is the starting market. Hotel rooms, flights, and broader participation in the economy are the expansion McCarty envisages, rather than capabilities established by the launch example.

An agent that buys something on Amazon makes an impressive demonstration: people applaud and repost it on Twitter. But McCarty points out that the merchant still pays Amazon, and Amazon receives credit for the transaction. Completing a purchase does not automatically give the agent builder either attribution or compensation.

Ionic's proposed alternative connects the enriched data to the purchase journey. Instead of receiving only a checkout result, a merchant would be able to understand:

• Data distribution: What product information went out into the world.
• Model use: Which model saw that information and how it used it.
• Action flow: How the interaction progressed to a transaction.
• Customer relationship: How the relationship continued after checkout.

This visibility is the economic counterpart to the shared decision-and-transaction layer. It would make the agent's role in generating a sale legible to the merchant, rather than leaving the merchant with only a marketplace conversion.

McCarty argues that merchants would prefer to pay builders for that contribution. He also acknowledges that Amazon pays, while claiming it pays less. The presentation supplies no matched commission terms for that comparison; the mechanism being proposed is compensation for attributable customer matching and transactions.

At launch, McCarty reported hundreds of merchant relationships available to builders immediately and a catalog containing millions of enabled SKUs. Those are Ionic's reported launch figures, without a breakdown of catalog coverage or checkout eligibility.

He closes by describing the catalog as AI native and continuing to grow, inviting prospective hires to contact Justin or Owen at Ionic Commerce, and directing builders to docs.ioniccommerce.com to get started. The invitation brings the three parts together: use richer product information to make a decision, carry that decision into a transaction, and give merchants a reason to fund the agent that helped.