Building networks for agents
The a16z crypto report explores how autonomous agents could power new two‑sided marketplaces for complex goods and services. It finds that agent‑driven networks can unlock transactions that are currently too costly or cumbersome for consumers and businesses.
ประเด็นสำคัญ
- Agents can translate nuanced preferences into machine‑readable requests, enabling automated matching of buyers and suppliers.
- Network effects similar to Uber’s ride‑hailing model would make agent marketplaces more valuable as participants grow.
- Founders can start with a narrow vertical—like restaurant bookings—and expand to travel, procurement, compute, and beyond.
- Revenue could come from order‑flow style fees, where the agent captures the spread between buyer budgets and supplier offers.
- Trust and reliability mechanisms (authorization, escrow, dispute resolution) are essential for scaling agent networks.
The Core Question
The a16z crypto brief asks whether autonomous software agents can replicate the network effects that made platforms such as Uber successful, but for far more intricate transactions. While Uber matches a rider’s simple request for a ride with a driver’s availability, the report envisions agents handling multi‑dimensional preferences—time windows, budgets, dietary restrictions, and more—across a wide range of markets.
Method and Conceptual Framework
Rather than presenting empirical data, the report builds a conceptual model. It draws a parallel to existing crypto‑based intent systems like UniswapX, where users post orders and liquidity providers compete to fill them. By inserting large language models (LLMs) to translate natural‑language conversations into structured intents, agents could automate the search, negotiation, and execution phases of complex purchases. The authors illustrate the idea with a dinner‑for‑six scenario, showing how personal agents could aggregate group preferences and solicit offers from restaurant agents.
Main Findings
The report concludes that agent‑driven marketplaces could generate powerful network effects. Each new restaurant or supplier adds a matching opportunity for more buyer groups, and each additional buyer group creates more demand for suppliers, reinforcing participation. It also highlights a viable business model: agents could operate for free to users while capturing the price differential between the buyer’s budget and the supplier’s quoted cost, akin to payment‑for‑order‑flow. This fee structure would fund the computational overhead of running LLMs and maintain user trust through transparent pricing.