Moves in related industries to develop software for AI agents, rather than for people, are taking shape.

Until now, AI agents have been trained to use software in the same way humans do. Against that backdrop, some companies are building software designed specifically for agents.

Axios reported on May 5 that Anthropic launched the Model Context Protocol (MCP) and presented a standard that connects AI systems with tools and data sources.

Salesforce last month announced the Headless 360 platform that allows agents to directly access core software. Stripe, Mastercard and OpenAI are adding payment infrastructure for agent shopping.

Parallel Web Systems, an AI agent tools startup founded by former Twitter CEO Parag Agrawal (파라그 아그라왈), is developing a web search and research API dedicated to AI agents. It counts Clay, Harvey, Notion and Opendoor among its customers. Parallel said banks and hedge funds are also among its clients but did not disclose specific names.

Zapier has focused for more than 10 years on breaking down software functions into unit tasks that agents can call directly, in order to automate them.

Zapier CEO Wade Foster (웨이드 포스터) said, "We are heading to a world where agents become the dominant software user." He added, "It does not mean people stop using software, but that there will be far more agents."

Box CEO Aaron Levie (애런 레비) said, "If agents become the biggest users of software, all software must be delivered in a headless way." He added, "Agents will not use a UI and will communicate through APIs."

Axios said that if agents stop using software like people, the core of the technology race could shift from who has the best interface to who controls the APIs, data and permissions that agents need. Former Meta AI chief Yann LeCun (얀 르쿤) said in an interview with Axios, "Agents will find ways to communicate directly and efficiently, without mimicking button clicks."