Big Tech pitches AI agents as the next major digital interface

Overview

Big Tech has a new pitch for how you will use the internet: you mostly won't. An AI agent will do it for you.

The Personal Agent Protocol

The most concrete move so far came on October 6, 2026. Meta and Sierra jointly announced the Personal Agent Protocol, an open standard for how personal AI agents deal with businesses. The protocol targets a basic problem. If your agent is going to buy something or change a booking, the business on the other end needs to know who sent it and what it is allowed to do. To handle that, the Personal Agent Protocol relies on OAuth for session management. Consumers decide what permissions their agent gets. Businesses define which actions an agent can take on their side.

The Efficiency Argument

There is also an efficiency argument. Today, many AI agents interact with websites the way a person would, by scraping screens and filling in forms. The Personal Agent Protocol aims to cut the latency that comes with that approach.

The Future of Agents

OpenAI made its own move around September 29, 2026, launching what it calls "dots," a set of always-on agents. They run on GPT-6 Astra, the company's model powering the product.

The Design

The design leans heavily on oversight. The agents are built to operate with user supervision and require explicit consent before carrying out high-risk operations. ChatGPT Work/Codex has over 35 million weekly users, a base large enough that a single misfired action could become a headline. Google is pushing its Gemini agent with a focus on enterprise workflows. Microsoft and Anthropic have their own agent initiatives running in parallel.

The Interface Fight

No single interface, whether chat, voice or browser, is expected to win outright. Instead, hybrid setups are emerging, along with agent-to-agent protocols. In that world, your agent negotiates with a retailer's agent, and you only see the result. That represents a real break from the graphical user interface. The priorities shift toward delegation, auditability and context-aware orchestration, meaning agents that know what you want, can show their work, and can coordinate multiple steps across services.

The Impact

For businesses, the calculus is different. A protocol where companies define permitted actions gives them more control than a world where agents simply scrape their sites. Retailers and payment firms signing on suggests they would rather shape the system than be scraped by it. Safety controls could prove to be the deciding factor. OpenAI's explicit-consent requirement for high-risk actions and the protocol's user-defined permissions both point to the same reality: people will hand over tasks only if they trust the agent not to overstep. For consumers, the near-term change may be subtle. Agents will appear inside familiar tools, handling bookings, orders and routine requests in the background. Whether that happens depends on interoperability. An agent that only works inside one company's walled garden is a feature. An agent that can securely deal with any business through a shared standard starts to look like an entirely new interface.