The end of ads? AI agents are about to change how we buy

The increasing capabilities of AI models, now often referred to as agents, are rapidly transforming how humans interact with computers, particularly in the realm of commerce. These agents, having advanced significantly in the last five months, are capable of understanding complex tasks over long time horizons and utilizing various tools to complete them. The key difference between a traditional Large Language Model (LLM) and an agent is that an agent is essentially a chatbot that can use a computer on your behalf, achieving parity with average human computer usage but at a significantly lower cost. This means anything a human can do with a computer, an agent can also do. This shift presents two main categories of agentic commerce. The first is "conversational commerce," where LLMs like ChatGPT facilitate checkout processes, offering empathetic recommendations and streamlining purchases. This is expected to be beneficial for consumers, merchants, and platforms alike, leading to better product discovery, increased conversion rates, and new revenue streams. The second category involves agents with more autonomy, capable of delegating money and paying for services to enhance their productivity for their human users. This is a "less skeuomorphic" version, suggesting a fundamental change in the internet's structure, moving towards an "agent-native internet" where agents directly pay for necessary resources. The long-term implications of this technology are profound. If an agent can gather all the information and perform tasks traditionally done by an app, the need for complex user interfaces might diminish. Users could simply instruct their agents to handle tasks like reading reviews, making purchases based on preferences, and managing logistics, effectively making non-programmers into "just-in-time natural language programmers." An agent internally writes and executes a program to fulfill these complex requests, often for mere cents in tokens and API calls, and then discards it, making such sophisticated actions accessible to everyone. One significant consequence of this shift is the emergence of the "headless merchant." Historically, developers would engage with enterprise sales teams to acquire tools and APIs, often involving lengthy onboarding processes and subscriptions. However, with agents, the human developer's intent is on what to build, not the specific resources used. This necessitates a commerce model where access to tools is on a per-usage basis, without the friction of enterprise sales or long-term contracts. A headless merchant, therefore, would primarily offer an API endpoint with excellent documentation, catering directly to AI agents rather than human navigation through a website. This model prioritizes speed and efficiency, allowing agents to quickly access and pay for services, thereby accelerating development and innovation. The elimination of friction in these transactions, while beneficial for speed and testing, challenges traditional business models that rely on lock-in and loyalty. However, new forms of stickiness are expected to emerge, such as reputation, memory, and data, as agents will likely favor services that have performed well in the past. This also implies a more efficient market, where agents, unswayed by sales tactics, will choose products based on performance and pricing, potentially disrupting established players who rely on expensive sales teams rather than product quality. This could democratize access to tools and services, creating a new market for API consumers who have no prior experience with APIs. The economic contract of the web, largely based on advertising and distraction since the early 2000s, is also facing disruption. Agents do not get distracted by ads, meaning the traditional revenue model for publishers will be undermined. A new economic contract will be needed, possibly involving direct payments for content or API resources. The question of "whose representative" the agent is becomes critical here. If an agent works for a platform like Google, it might still be subject to the platform's interests and potential "distractions" (ads). However, if an agent is controlled by the user and is open source, it can be equipped to avoid such influences. The debate also extends to payment rails. While credit cards are suitable for traditional e-commerce due to baked-in consumer protections, they are less ideal for the microtransactions characteristic of agentic commerce due to fixed transaction fees (e.g., 30 cents per transaction) and month-end settlement. Stablecoins, with their sub-cent fixed fees and instant settlement, offer a more suitable alternative for these small, frequent transactions, as they eliminate the need for merchants to extend credit to unknown agents. While credit card companies could theoretically adapt by introducing microtransaction-specific pricing, their existing business models and the psychological appeal of rewards might make this unlikely. However, the increasing ease of acquiring and using stablecoins, coupled with banks offering native stablecoin support, suggests a future where stablecoins play a significant role in agentic commerce. Ultimately, the mass adoption of agentic commerce will be driven by convenience. Agents can perform complex tasks, like group ordering groceries, building CRMs, or managing subscriptions, far more efficiently than humans. The current, often convoluted, processes for managing online life, from subscriptions to financial transactions, are ripe for disruption. A sufficiently capable agent will simply be better at these tasks, offering a more streamlined and integrated experience where information and transactions are seamlessly connected. This new paradigm of "composable commerce," where agents can stitch together various APIs and instantly settle complex, multi-party transactions, promises a more efficient and lean ecosystem for businesses and consumers alike.