David Pawlan, creator of Assistant Benchmark, stated on the Andreessen Horowitz podcast that after the internet, there will be infrastructure and products specifically for interactions between agents; services will evolve from person-to-person bookings and agent-to-person bookings to agent-to-agent bookings, rewriting the entire service industry.

General Partner Anish Acharya first proposed that infrastructure and products specifically for agent interactions, which do not exist today, will emerge. Pawlan added forms such as agent email and agent phone numbers, and questioned how the service industry will define itself when "there are no direct orders from people." He outlined the sequence: first the internet, then cybersecurity, first agents, then security components for agents, and questioned what industries that would spawn. The conclusion is that there are enough niche problems for a batch of startups to build a new layer of digital connections.

This judgment comes from the same 51-minute discussion, where Pawlan tested booking tickets and finding restaurants using unified prompts. Agent interactions mean that merchant interfaces need to change from pages that serve people to machine-readable channels that provide software credentials, limits, and audits. Email and phone numbers are considered legitimate identities for agents, rather than one-time conversations in a chat window. Security is singled out as a new industry, addressing issues like unauthorized orders, impersonating merchants, and collusion between agents to lock inventory.

On the same day, the enterprise side has already integrated agents into market shelves and cloud environments; this speech pushed the next step from "clicking buttons for people" to "two pieces of software settling accounts with each other." No new evaluation champions or funding amounts were announced, only predictions about industry structure.

In market mechanisms, what is bought is the yet-to-exist agent identity and interaction protocols, while what is sold is the expectation that the intermediary layer of the service industry will be rewritten. The funding narrative points to startups in identity, payments, inventory, and security, rather than creating another chat assistant; beneficiaries are platforms that can issue agent credentials, while those under pressure are booking phones and web forms that only serve human clicks. The event driver is podcast snippets, not standard releases.

In person-to-person service industries, charges are based on seats, while agent-to-agent charges are based on interfaces and success rates. Once emails and phone numbers are given to software, the cost of impersonation decreases, and the security industry can then have independent billing.

Source: Public Information