# When Your Customer Is an Agent

**Track:** Building the AI-Native Product — AI for Entrepreneurship — complete (29)
**Framework / surface:** venture strategy
**Level:** Advanced
**Prerequisites:** Workflows vs. Agents

**In one line:** Machine-readable capabilities, delegated auth, meterable pricing — designing for the buyer who never sees your landing page.

## Theory, aesthetics & inspiration

Software's newest buyer never sees your landing page: an agent, acting on a customer's behalf, discovers capabilities from machine-readable descriptions, negotiates authentication and permissions, compares prices, invokes the tool, and judges the result. The infrastructure for that world is consolidating fast — Anthropic donated the Model Context Protocol to the Linux Foundation's Agentic AI Foundation in December 2025, with OpenAI, Google, Microsoft, and AWS participating, and the July 2026 specification revision pushed the protocol toward stateless, enterprise-scale agent deployments. For a founder this is a second product surface with its own go-to-market: capabilities must be described the way an agent parses them, permissions delegated safely, pricing meterable per invocation, reliability legible as machine-checkable trust signals rather than testimonial logos. Every question this track asks about human customers re-asks itself mechanically — how do agents discover you, what convinces one to choose you, what makes it return? The companies that answer early become the suppliers the agentic economy routes through by default.

**Founder question:** When an agent evaluates your product on a customer’s behalf, what does it see?