In October 2026 I reflected on business models for AI agents after a question from Michael Mignano at USV about who pays for an agent that works at 3 am. I run The Residency Bangalore, a program backing deep‑tech founders in India, and I explored how free and paid tiers might coexist.
- Every AI‑agent task incurs inference cost.
Decisions: 1. Adopt an advertiser‑sponsored model for free agents, where sponsors pay to have their answer displayed. 2. Offer both a paid tier (direct user payment) and a free tier (advertiser‑backed) to capture the large Indian user base that has never paid for a subscription.
Methods include defining “agent ads” as the answer output, building a marketplace for sponsors to bid for answer selection, and testing the model first in India where most users have never paid for a subscription.
- Open questions:
most of our billion internet users have never paid a subscription for anything.
- How to price advertiser bids to ensure sustainable revenue while keeping answer quality high?
- Which metrics decide when to offer a paid‑agent upgrade?
- How to address regulatory and privacy concerns around advertiser‑influenced AI answers?
- Can this ad‑based model scale beyond India into markets with higher subscription adoption?
- Charging per task is logical but falling inference prices risk a race‑to‑the‑bottom.
- Pay‑per‑outcome sounds clean but most tasks are too small to monetize individually.
- Historical “free” internet services (Google, Facebook, Jio) were funded by advertisers; users paid with attention.
- AI agents can adopt the same model: an “agent ad” is the answer the agent selects for the user.
- I anticipate a market split: a paid agent that works directly for the user, and a free agent that works for whoever paid to be chosen.