In the evolving landscape of software procurement, AI agents are emerging as significant players, yet a noticeable gap exists in how they interact with SaaS applications. Currently, when an AI agent attempts to evaluate or utilize a software tool, it relies on scraping marketing pages in a manner reminiscent of the early internet days. This outdated approach highlights a lack of standardized methods for pricing discovery, understanding product functionalities, and completing transactions without human intervention, which can disrupt the entire process.

To address these challenges, three promising solutions have emerged:

  1. llms.txt: This is a simple text file placed at the root of a domain that provides AI agents with essential information about pricing, policies, and product capabilities. It functions similarly to robots.txt but is tailored for Large Language Models (LLMs). Although the specification is available, adoption remains limited.

  2. MCP servers: These servers allow developers to expose core functionalities of their products as callable tools. This enables AI agents to perform actions like list_plans() or create_project() directly, streamlining their interaction with the software. The specification exists, yet many SaaS offerings have not implemented it.

  3. Agent checkout protocols: Systems like ACP facilitate agent-led purchases without human-like confirmation screens or redirection flows, allowing for a smoother transaction process.

The pressing issue is that as more research and decision-making shifts towards AI agents, companies risk losing out on potential sales if their products are not easily discoverable or assessable by these non-human entities. Have you noticed any increase in agent traffic within your analytics? Have you adopted any of these solutions, or are they still on your radar? Would you consider investing in a service that manages this aspect for you, or is it a task you’d prefer to keep in-house?