The agent economy is getting a cash register

Loomal’s pitch is wonderfully blunt: monetise an MCP server. Not an app, not a SaaS seat, not a dashboard with a pricing page bolted on later. A server. A tool. A thing an agent might call in the background while a human barely notices.

That is the detail to sit with. The agent economy is getting a till.

The obvious reading of this week’s stories is that AI companies are searching for revenue. True, but too shallow. The deeper shift is that AI products are being rebuilt around commercial intent: pricing, billing, attribution and proof that something valuable happened. The old question was “what can the model do?” The new one is “who pays when it does it?”

OpenAI is testing ads inside ChatGPT, which raises the trust problem for assistants. If the assistant that answers your question also has paid suggestions in the room, people will want to know where the answer ends and the advert begins.

But the more interesting product question is whether conversation becomes a distribution channel. Search advertising worked because intent was legible: someone typed “running shoes”, and the auction knew what to do. ChatGPT has something stranger and potentially more valuable: a conversation that can contain goals, constraints, preferences and half-formed plans. If ads enter that surface, they are not competing for attention in the same way banner ads did. They are competing to become part of a decision.

That makes separation between assistant output and commercial content more than a UX nicety. It becomes the trust boundary of the product.

The new rails

Loomal’s Product Hunt page points at the other side of the same market. If agents are going to use tools, APIs and digital products on a user’s behalf, those services need paywalls that agents can understand. Human checkout flows were designed for people: plan pages, card fields, confirmations and invoices. Agents need something closer to a machine-readable commercial contract.

This is where the hype around agents gets less cinematic and more useful. Nobody wants to watch an agent browse a pricing page and guess whether a tool call is worth it. Builders need permissions, prices, metering and receipts built into the workflow. Loomal is a small launch, but the category it gestures towards is not small.

UnitPay’s Product Hunt listing makes the pain more explicit. It describes itself as a monetisation OS for AI companies, covering pricing, usage-based billing, inference costs and margins. That last part matters. AI products do not have the clean cost profile of old SaaS. A user clicking one button may trigger model calls, tool calls, retrieval, image generation, evaluation and follow-up actions. The product team sees a feature. Finance sees a meter spinning.

This is why pricing is becoming a product engineering problem. If you cannot connect usage to value, you either overcharge good users, subsidise expensive ones, or hide the whole mess inside vague subscription tiers until margins punish you.

Creatify’s AI Media Buyer completes the loop. Creatify is moving from AI video ad creation into AI-managed media buying, promising ad operations that improve over time. Put that beside OpenAI’s ad tests and the pattern gets sharper: AI is becoming both a place where ads may be sold and a tool for deciding how ads are made, placed and measured.

The retail parallel is the boring machinery behind the shopfront: tills, receipts, stock systems and card networks. Those systems did not make shopping glamorous. They made it accountable. AI is going through the same phase. The demo layer produced magic; the commerce layer has to produce settlement.

I think this is the next serious bottleneck for AI builders. Agents will not become economically meaningful because they can click buttons. They become meaningful when they can make priced decisions inside trusted limits, and when every party can see what was bought, why it was bought and whether it paid off.

The winners may not be the flashiest agents. They may be the products that know exactly when to ask for money, how to justify it, and how to stay trusted after the invoice lands.

---