No Humans

The web is changing its nature. In just a few years, automated traffic has surpassed human traffic, and a new layer of protocols is emerging to allow artificial intelligence agents to act on our behalf. Welcome to the agentic web.

The tipping point: humans are no longer the majority

As of June 2026, according to Cloudflare data, AI agents account for 57.4% of global web requests, compared to just 42.6% for humans. This tipping point, which many anticipated for the end of the decade, is already here.

The trend is steep. In 2024, automated traffic already reached 51% according to Imperva’s Bad Bot report ( Imperva was later acquired by Thales). What has changed is the very nature of bots: malicious robots (scrapers, attacks, spam) are no longer alone in dominating. AI agents — bots that navigate, understand intent, and execute tasks — saw their traffic increase by over 7,000% in one year.

This is no longer a simple replacement of humans by scripts. It is the emergence of an ecosystem where autonomous software entities interact with web services designed for and by humans.

What is the agentic web?

The agentic web refers to this new layer of the web where AI agents no longer simply answer questions — they act: they search, compare, buy, book, negotiate — often without direct human intervention.

Unlike traditional chatbots, these agents understand the intent behind a complex request. A user no longer says “show me camping tents,” but rather “find a camping tent under €150 delivered by Friday, preferably lightweight for mountain hiking.” The agent interprets the constraints, queries multiple catalogs, aggregates offers, and proposes — or executes — the best option.

This autonomy rests on three pillars:

  1. Intent understanding: the semantic agent goes beyond keywords to grasp implicit constraints (budget, deadlines, contextual preferences).2. Multi-platform action: the agent can interact simultaneously with multiple services, without the user having to navigate from site to site.
  2. Autonomous execution: adding to cart, calculating fees, finalizing payment — all via secure protocols, often without even redirecting the user to the merchant’s site.

The protocols building the agentic web

For this ecosystem to work, new standards are emerging, designed specifically for interoperability between agents and platforms.

Universal Commerce Protocol (UCP)

Proposed jointly by Google and Shopify, the UCP standardizes access to product data and the initiation of transactions. Concretely, it allows any AI agent to “understand” any online store without specific integration. It is a bit like the HTTP protocol that allowed any browser to access any web server.

Model Context Protocol (MCP)

Developed to allow agents to retain and share user context — preferences, history, authentication — across different applications, MCP solves a fundamental problem: the fragmentation of digital identity. With MCP, your shopping agent can share your size preferences with your travel agent, which already knows you prefer downtown hotels.

Agent Payments Protocol (AP2)

Financial transactions require a framework of trust. The AP2 protocol, driven notably by Visa with its tokenized credentials, enables secure payments with delegated authentication. The agent pays on your behalf, but with security guarantees and traceability.

The players already on the field

This is not science fiction. The biggest digital players have already launched concrete features:

  • Amazon has deployed the Buy for Me program, where an agent can purchase products outside the Amazon marketplace on behalf of the user.
  • Shopify launched Agentic Commerce in August 2025, allowing conversational AI assistants to purchase directly from their interface on Shopify platforms.
  • OpenAI has equipped ChatGPT with a genuine agent (launched in July 2025) that has a virtual computer to navigate, fill out forms, and make purchases online.
  • Stripe offers tools via the MCP protocol to integrate payment services directly into AI agent workflows.
  • Tencent announced the development of an AI agent within WeChat during its earnings call, signaling that the concept is now extending to Asian mobile platforms.

What this means for merchants

For retailers, this evolution imposes a paradigm shift. AI agents do not read banner ads or get influenced by attractive design. They evaluate structured data: availability, verified reviews, consistency of information.

This is the emergence of Generative Engine Optimization (GEO) — optimization for generative engines. A product’s visibility will no longer depend solely on traditional SEO, but on the quality and consistency of its structured data. A price that changes between two interfaces, an inconsistent description, poorly synchronized stock — and the agent silently eliminates you.

Data from MasterCard and IBM confirms this trend: merchants who do not properly structure their product data will be invisible in the agentic era.

Open questions

This shift raises fundamental questions:

  • Who is liable when an AI agent makes an unwanted purchase? The agent developer, the merchant, the user?
  • How do we ensure competition if agents systematically favor certain partners through commercial agreements?
  • What remains of the user experience — that browsing, serendipitous discovery, digital wandering that has been part of the web since its beginnings?

What is certain is that the web designed for humans — with its pages, clicks, its visual interfaces — is mutating toward a web designed for agents. A web of structured data, standardized protocols, and automated actions.

The glass is almost full. And it is no longer the human holding it.

Sources: