The Era of Traditional E-commerce is Coming to an End
The evolution of Artificial Intelligence agents is driving a disruption by replacing platforms with autonomous connected systems.
Credit: Article by Diego Barreto, originally published in MIT Technology Review Brasil.
Since lockdowns forced consumers to shift their purchasing online, e-commerce has undergone immense evolution. The focus was on three pillars: inventory (or effective stock control), payment, and delivery. Customers searched for products using search engines like Google and were directed to platforms where they completed the purchase. The central proposition was based on one word: convenience. But that is changing. And at an accelerating pace.
With advancing technology, the user experience, which previously revolved around platforms, began to be carried out directly with Artificial Intelligence agents, or AI agents, as they are known. In a first phase, these solutions emerged to improve the experience within the platforms themselves and functioned as layers of assistance. Amazon, Booking.com, and OLX began incorporating agents to personalize recommendations, facilitate natural language searches, and anticipate consumer intentions.
The next step was to take over the top of the sales funnel, a phase where the main goal is to spark customer interest. Tools like ChatGPT and Perplexity have become alternatives to traditional search engines, processing the search intent, navigating options, and presenting the best decision directly, rather than redirecting the user to a marketplace.
What does this change for e-commerce?
This movement is important because it means e-commerce sites cease to be the starting point for the shopping experience. The relevance of a brand, in this new scenario, depends on how integrated it is with the systems connected to the agents. And the work is not limited to attracting the consumer: agents help convert the buyer into a repeat customer. They analyze data, make decisions, execute tasks, and learn from the results. It is a never-ending cycle of improvements, with agility previously unattainable.
This is clear when we take a restaurant registered on the iFood platform as an example. Imagine it wants to know details about a recent drop in sales. It simply makes the request via WhatsApp, through Salesforce, to the AI account manager. From there, this agent contacts the AI data analyst, who inspects what is happening, identifies the problem, and offers recommendations. The AI account manager receives this data, analyzes it, and adjusts the recommendations before sending the response to both the restaurant and the iFood account manager.
Looking Ahead to the Future
The next stage in agent evolution is directly related to language standardization. Just as HTML was essential for the web, the Model Context Protocol (MCP) emerges as the infrastructure that will allow this technology to connect to multiple marketplaces fluidly.
It is an open-source standard created to interlink AI models with data sources, replacing fragmented integrations with a single protocol. It is a simpler and more reliable way to provide systems with access to data, making them more powerful and scalable.
With MCP, an agent can search, compare, transact, and execute complex tasks with different suppliers without relying on a specific interface. Interoperability becomes the new battlefield, replacing the war for traffic with a war for integration.
How it Works in Practice
The concept has already become a reality. Shopify, one of the world’s leading e-commerce platforms with over a million merchants, has enabled MCP across all its stores. In recent weeks, OpenAI and Stripe partnered to launch the Agentic Commerce Protocol (ACP), an e-commerce protocol featuring instant checkout, allowing purchases to be made directly through ChatGPT or Perplexity.
The user asks ChatGPT about a product and discovers the best offers compatible with their profile available on the web because the chat itself suggests the options, becoming a storefront. They simply tap the “buy” button for the commercial operation to be conducted entirely within that interface, without redirects. In this first phase, instant checkout is being implemented in the United States for purchases from Etsy and Shopify marketplace sellers.
Window of Opportunity
The development of agents has been extremely fast, and their use is already showing results. According to McKinsey research, companies that implemented service agents managed to increase problem resolution by 14% and reduce the time spent dealing with them by 9%. Remember: these numbers reflect only the initial thrust of the revolution to come.
According to the global research and consulting firm Gartner, agents will be at the forefront of the next great technological wave. According to the Hype Cycle for Artificial Intelligence, by 2028, 33% of corporate software applications will include generative AI, with at least 15% of daily work decisions made autonomously through agents.
In a more advanced scenario, these assistants will not only interact with humans but also with each other. In the future, which is actually much closer than we usually think, we will have autonomous ecosystems where consumer agents and supplier agents negotiate, transact, and optimize workflows without human intervention.
When Humans Exit the Scene
This “A2A” (Agent-to-Agent) model paves the way for a new platform paradigm: they cease to be intermediaries and become execution environments. The most emblematic example is the Vend project, presented by Anthropic, in which a vending machine was 100% managed by an agent, from inventory to pricing, communication, and acquisition of supplies.
Named Claudius, Anthropic’s agent assumed the manager role for a month with the mission of making a profit, predicting demand, searching for products to buy from wholesalers, defining prices passed on to consumers, and handling customer service, among other inherent business tasks.
The result was not the best: Claudius got some things right but failed in many others. What matters in this story is seeing that AI is very close to performing this type of work without any supervision. It shouldn’t take long for this technology to be perfected.
Difficulties Along the Way
It is worth remembering that the autonomy of these systems brings a significant challenge. By interacting with different environments and executing tasks independently, agents create regulatory risks associated with decision-making and the capacity to act in the real world. Companies need to implement robust encryption and secure storage practices to protect data privacy and security.
And it’s not enough just to create a network of integrated agents. It is necessary to ensure reliability, understand what they can and cannot do, and how they should be governed. Agents need testing, training, and guidance before operating independently.
Capability and Trust
Again taking iFood as an example, the reliability and capability of the company’s agents have been evolving at a predictable and very accelerated rate. The complexity of tasks that the models can successfully perform doubles every seven months. How is this possible? With strategy, investment, culture, and a lot of testing to advance step by step.
The company maintains a program that has reached 880 active agents per week, created by employees from different areas. The goal is that, by March 2026, this number will reach 7,000 agents, one per employee. Each employee is at the center of the process, with the autonomy to create tailor-made assistants that reduce work previously done in days to mere seconds.
The New Order of Commercial Internet
The conclusion is clear: for companies operating in e-commerce, value will no longer be in the interface but in the ability to connect to a network of agents. This changes everything. Technological priorities will be different.
Building APIs will cease to be sufficient, and the new competitive edge will be becoming **“agent-ready”**, which means being prepared for the internet era where Artificial Intelligence will be able to understand the environment, make independent decisions, perform complex tasks, and adapt to needs in ways we cannot yet imagine. Human employees will become the architects of these ecosystems.
Platforms that quickly adapt to this reality, offering MCP-compliant interfaces, structured data, and accessible transactional functions, will be prepared to prosper in the new digital order. For brands, the game also changes. They must be attentive to a new consumer: autonomous software that makes decisions on behalf of another person.
The agent revolution is not just a technological evolution but a reconfiguration of the commercial internet architecture. From interface-based e-commerce to autonomous negotiation-based e-commerce, we are migrating from a user-centric world to an AI-centric world.
The future will not be made of platforms competing for our attention, but of protocols fighting for our integration.
