YC dropped a thesis this week: the next trillion internet users will be AI agents (Agents: digital employees capable of autonomously executing tasks). This means all software designed for human "clicks and taps" needs to be rewritten for machines.
What this is
YC's core logic is clear: current software interactions—buttons, menus, forms—are entirely designed for human brains and fingers. AI agents are currently forced to simulate humans clicking interfaces, which is highly inefficient and error-prone. YC believes every major software category needs reconstruction to make agents "first-class citizens," meaning skipping visual interfaces and completing tasks directly through APIs (Application Programming Interfaces, the channels for software-to-software dialogue) or underlying protocols. YC asserts that this reconstruction won't be done by incumbent software giants because they are constrained by existing human-interaction paradigms and commercial baggage.
Industry view
We note that YC's statement is essentially a VC issuing a "rallying cry," seeking new entrepreneurs to reconstruct software. "Incumbents can't do it, only new players can" is a classic VC narrative with its own rationality: modifying systems at big tech firms has cascading effects, making them genuinely slow to act. But the risks here are worth our concern: critics argue that incumbents aren't sitting idle. Microsoft, Salesforce, and others are integrating agents into existing systems via plugins and Copilot; furthermore, agent protocols are currently highly fragmented. Once incumbents unify standards at the OS layer, startups could easily be relegated to plugins within an ecosystem rather than remaining independent platforms.
Impact on regular people
For enterprise IT: Software procurement logic will shift from "is it easy for employees to use" to "is it easy for machines to interface with," and the core criterion for evaluating SaaS tools will skew toward API openness.
For individual careers: Execution-level work involving manipulating software interfaces will be rapidly replaced. Human professional value will shift upward to "defining tasks" and "auditing results."
For the consumer market: The short-term experience won't change much, but once the underlying switch is complete, individuals in the future may no longer need to subscribe to multiple software applications, but rather hire a universal agent to operate on their behalf.