Dissecting how AI agents can expand Uber-style matchmaking to complex transactions, pointing out that bilateral networks and trusted transaction mechanisms are the real moat.

Written by: Guy Wuollet

Translated by: AididiaoJP, Foresight News

Uber simplifies demand into one action: calling a car from A to B. The network is responsible for finding drivers willing to accept orders at a reasonable price. The more drivers there are, the easier it is for passengers to get a ride; the more passengers there are, the more willing drivers are to join.

The author believes that AI agents can extend this matchmaking network to more goods and services. Agents can clarify complex preferences on behalf of people and then seek willing suppliers. Some transactions exist today, but prices are too high or processes are too convoluted for ordinary people to participate; some transactions are too troublesome for anyone to organize; and there are additional demands that people haven't even thought to raise. For entrepreneurs, these are all opportunities.

Can an Uber-style network be used to match a meal?

Getting friends together for a meal is a familiar challenge. Everyone has different schedules, budgets, and tastes. Typically, one person is responsible for collecting preferences in the group and searching for restaurants; if the options are not suitable, they have to go back to the group for further discussions. On the restaurant side, there may be empty tables available on a Tuesday night, but it's difficult to reach a group that hasn't finalized their plans.

Agents can take over this task. Your agent would communicate with friends, or coordinate with friends' agents, finally forming a request: six people, in the city, no more than $60 per person, anytime between 5:30 PM and 8:00 PM, eating everything except sushi. The restaurant agent quotes based on table availability. One restaurant might say that the price is better for a 5:30 PM seating; another might offer a family-style menu for this group. Your agent matches the quotes with everyone's preferences and places an order within everyone's agreed range.

If a restaurant has available tables at six o'clock, it can directly quote the group willing to eat early, even before they have selected a restaurant. The group has preferences and a budget; the restaurant has capacity and acceptable prices. What’s missing is a network that allows both parties' agents to find each other and agree on a purchase.

Uber has proven that a network can bring buyers and sellers together around relatively simple demands. Agents can expand this model to accommodate more complex purchases, such as a dinner for a group of people. Most networks are built for humans; it is now time to build for machines. Entrepreneurs can create markets like this: agents come in with requests, and suppliers compete to fulfill them.

A market for "machine use" between agents

The cryptocurrency field already has models for organizing such requests: intents, which are intention orders. Users only specify the desired outcome, and suppliers compete to deliver. For example, with UniswapX, users sign an order indicating what tokens they want to exchange and under what conditions; suppliers compete to match requests, sourcing liquidity from different places.

Agents can extend the intent model to everyday commerce. Large language models can transform a conversation into requests actionable by suppliers, and agents can then search for quotes and negotiate details. Arranging a relatively small purchase, such as a dinner for six, could be cheap enough to justify this approach.

Uber is useful because of network effects. More complex purchases can also generate similar network effects. With each additional restaurant joining, diners gain another matching opportunity, especially when groups have specific requirements. The more groups there are, the more chances restaurants have to fill their empty tables, making it even more worthwhile to integrate reservation systems and provide tailored quotes. The more participants in a city, the broader the range of requests the network can handle.

Entrepreneurs can start with restaurant services in one city and then expand. The initial product must save users enough time or money for them to want to bring friends on board; restaurants need to see bookings they would otherwise miss out on. Only when both sides achieve results will they return and invite others.

First, let both sides benefit

The author believes that consumer-facing agents can be used for free, charging through an order flow commission model. Users specify what they want and how much they are willing to pay, while suppliers compete to take orders. If the agent agrees to arrange a dinner for $300 and finds a restaurant that can do it for $270, the network can retain the price difference; it could also charge the restaurant a referral fee. This transaction revenue can be used to cover the reasoning costs for users.

But users should know how agents charge and how they make choices between quotes. Whether they are willing to delegate purchases to agents depends on their trust in those choices.

Software can be copied, but networks and trust are hard to replicate

AI makes software increasingly easy to copy, and the willing participants in a network transaction, both people and businesses, become more significant sources of advantage. Some of the best companies of the internet era, like Facebook, eBay, and Uber, operate better with more users.

Agent commercial networks can also create such advantages. Competitors might produce similar software, but they still need to bring buyers and sellers together and ensure that transactions are reliably completed. Returning to the restaurant example, this means there must be enough local restaurants willing to quote, as well as enough groups ready to place orders.

Ensuring reliable transactions is part of building a network. Restaurants need to know that agents have the authority to access this group's funds; diners need to understand that accepting a quote locks a table at the agreed price. The network must be able to verify agent authority, record reservation terms, and manage payments. Both sides should also be aware of the consequences if the group cancels or the restaurant fails to fulfill the order.

Entrepreneurs can build markets while simultaneously establishing these services. As buyers and sellers transact, the network can accumulate records of who honors commitments and who fulfills orders, developing a dispute resolution process. A network that can help participants find good quotes and reliably complete purchases will encourage people to return for future transactions.

Not just dining: travel, procurement, computing power, and arbitration

Dining is just a toy example. Agents can help build many networks.

In travel, personal agents can coordinate a group’s schedules and budgets, negotiating suitable arrangements with hotels and other suppliers. The more flexible the dates or destinations, the more likely they can find quotes that fixed searches may not reveal.

In corporate procurement, corporate agents can be online all day, negotiating future supply and price commitments with suppliers. They compare not only prices but also delivery plans and payment terms, helping buyers lock in costs while providing sellers with more predictable demand.

In computing power markets, agents match workloads and available capacity based on hardware requirements, location, and time. Companies can even purchase future computing power to hedge against costs.

In arbitration, parties agree to resolve disputes using a shared arbitration platform, which is faster than existing legal systems. The more buyers and suppliers using the same set of rules, the more attractive the platform becomes to the next participant.

In the realm of physical device networks, agents can buy services from drone networks or data from connected devices. The more providers there are, the wider the network's coverage and the more types of services needed by other participants it can offer.

In collaborative health research, people can aggregate health data and participate in crowdsourced studies. The more contributors there are, the better they can support larger-scale research and answer questions that small groups cannot.

Human-centered networks also hold potential. As intelligence becomes cheaper, there remains a need for individuals to establish identity, endorse others' qualifications, and build trust relationships. With more work occurring outside traditional institutions, organizations like credential networks and member clubs may become even more valuable.