Aaron argues that we have reached a critical turning point where agents are replacing humans as the main users of software, even though the entirety of internet infrastructure—including protocols, web standards, pricing models, and enterprise software—was originally engineered specifically for people. This demands a complete overhaul of existing infrastructure, impacting everything from web standards and pricing models to how businesses build and deploy software, fundamentally redefining digital interaction.
This profound change is largely driven by a projected explosion in agent numbers. According to a recent post by Aaron on X, the number of agents operating across digital systems could soon outnumber human users by 100x to 1,000x, demanding a comprehensive reassessment and fundamental transformation of software development and deployment.
Adapting Paul Graham’s maxim, developers must now pivot to “make something agents want,” ensuring systems offer:
- API-first design
- Structured data
- Tidy codebases
- Current documentation so agents can operate effectively
Agents do not attend webinars or respond to outbound emails; they identify effective tools and integrate them directly.
The internet’s adaptation is unfolding unevenly across sectors. Business-to-business (B2B) companies, like Box, are ahead, having already released command-line interfaces (CLIs) that work well for agents and developed APIs for integration. This contrasts sharply with the consumer side, which faces significant challenges.
Consumer-focused businesses, digital publishers, and e-commerce platforms share a well-established, 20-year record of actively keeping bots out. Their revenue, particularly for publishers and e-commerce giants that rely on advertising, is tied to human eyeballs, creating a disincentive to quickly enable agents to browse or purchase autonomously.
For example, the inability to send an agent to buy items on Amazon or read a subscribed article from The Information or Wall Street Journal highlights the current friction. As Matthew Berman points out, this advertising revenue model means these companies are disincentivized from rapidly transitioning to an agent-first world.
Emerging agent technologies are pushing this boundary. Perplexity introduced “Computer,” a cloud-hosted agent that coordinates 20 distinct AI models to independently carry out complex, multi-stage workflows that can run for hours.
They followed this with “Personal Computer,” software designed to run persistently on a Mac mini. The viral success of Open Interpreter, an open-source agent built by Peter Steinberger, which acquired 200,000 GitHub stars and was subsequently acquired by OpenAI, further underscores this trend, sparking a wave of similar personal AI assistants from companies like Tencent and Cohere.
Within enterprises, Rob Seaman, EVP and GM of Slack, highlights how Slack, with 47 million daily active users, is positioning itself as an “operating system for the agentic enterprise,” leveraging its rich channel-based context to facilitate human-agent interaction and agent distribution.
How Autonomous AI Agents Work Through Recursive Loops and How Autonomous AI Agents Work Through Recursive Loops explore how agents function and integrate into larger systems. However, the lack of thorough process documentation within many enterprises presents a fundamental hurdle for implementing agents at scale, requiring significant internal preparation.
The Inevitable Shift to an Agent-First Internet
The future of AI agents is one of pervasive integration, fundamentally reshaping how we build and interact with the digital world. While the energy efficiency of computing has increased by 10,000 times, enabling more complex operations from models like GPT-5, Gemini-3, Llama-4, and Grok-4, the journey to an agent-first internet is fraught with technical and business model challenges.
The consumer internet, with its legacy of bot-blocking and ad-driven revenue, will be the slowest to adapt, but the momentum of agent innovation, supported by developments like Claude Code and Claude Co-work updates, suggests an inevitable shift towards systems optimized for autonomous interaction. What AI Agent Security Means for New Threats highlights critical considerations as this new field emerges.
