During an appearance on the a16z podcast, Greg Brockman, president and co-founder of OpenAI, contended that the software industry has spent considerable effort retrofitting applications to accommodate AI agents. He suggested that computer use capabilities could provide a more straightforward alternative to this approach.
Brockman emphasized that simplicity should guide AI development philosophy: "We really should be one AI that's unified, that makes it so easy and smooth for you to be less engaging with the computer and less wrapping yourself around the computer."
However, other organizations in the AI space continue to prioritize connector-based infrastructure for agent workflows.
The case against purpose-built integrations
Model Context Protocol servers, command-line interfaces, APIs and similar agent connectors enable AI systems to access tools, data and applications. Yet this accessibility requires ongoing engineering effort, as Brockman explained during his conversation with a16z hosts Ben Horowitz and Erik Torenberg.
"People have been building these MCP servers and these CLIs and just really sort of taking the world of software and making it accessible in this almost stilted way that is not really meant for humans," Brockman said. "It's like we're kind of retooling the world."
The proliferation of connectors stems from the requirements of agentic systems. Yet if AI models could operate computers as humans do, developers might avoid building custom integrations for each application. Brockman posed the question: "What if it's more behaving like a human? Can it just use a computer?"
Computer use has been part of OpenAI's roadmap for nearly a decade
According to Brockman, OpenAI's leadership outlined a three-step strategic plan during an offsite meeting in November 2015, which the organization largely followed over the subsequent decade.
During that same period, the team explored applying reinforcement learning to computer interfaces. Brockman recalled: "We also talked about, what if we could do reinforcement learning where the environment is screen pixels, keyboard, mouse, right? Same interface as a human."
From Brockman's perspective, this capability could unlock agent functionality across nearly any task. More significantly, he believes computer use could enable the industry to advance AI systems without requiring purpose-built integrations at every stage. Rather than creating and maintaining numerous specialized connectors, agents could leverage the same interface humans use.
"There's so much software that you don't even think about," he told Horowitz and Torenberg. "How much of your life is clicking around menus and typing things into a spreadsheet and things like that? None of that is what we should be doing."
Brockman credits computer use capabilities as part of why he considers GPT-6 Astra, OpenAI's latest flagship model, to be "pretty reasonable to call" artificial general intelligence.
Structured integrations remain central to industry strategy
Despite Brockman's enthusiasm for computer use, the broader industry continues to invest heavily in structured connector approaches. AWS recently unveiled a managed consent portal, managed web experience and session binding endpoint for AgentCore Gateway, a component of Amazon Bedrock AgentCore that facilitates agent connections to external tools and services. The company published detailed documentation of the feature on Monday.
The consent portal allows administrators to designate services such as GitHub and Slack as gateway targets, then distribute the portal URL to developers. Those developers can authenticate through their corporate identity provider and establish connections to GitHub and Slack independently.
OpenAI itself has not abandoned the connector model. Although OpenAI Codex arrived in browsers via a Chrome extension in May, enabling agents to operate within live browser sessions and authenticated workflows across multiple tabs, plugins remained the preferred integration method. This preference exists because plugins allow Codex to interact directly with services including Slack, Gmail and GitHub without requiring manual navigation of their user interfaces.
While Brockman maintains that simplicity should serve as the guiding principle for AI development, and reliable computer use could help achieve that goal, connectors continue to play a significant role in current agent architectures.