AI is a bad biz model but can be a great GTM for better ones - software, platforms, and marketplaces

AI agents suck as a business model bc they’re expensive to run. BUT they will be a clever hack to launch new marketplaces and platform and software businesses that wind up looking rather like their “Pre-AI” counterparts.

Agents talking to agents or people is dumb and expensive, but unavoidable; it shouldn’t the end goal and it doesn’t have to be. Instead LLMs ~should~ be a bridge to turn more the world into APIs and transaction layers.

Here are the steps:

1. Subsidize compute to offer a manual, outbound service cheaply with AI (making phone calls, eg), even if it means low gross margins

2. DDOS the other side of the market with tons of newly cheap, automated inbound. This is Jevon’s paradox for conversational AI.

3. Sell them “Cloudflare” to protect their human resources - an agent that manages all the newly induced inbound activity

4. A growing number of transactions will become closed loop - instead of agent to agent they’ll just be data exchange (which is much cheaper)

5. Then you can sell the data exchange, marketplace, or interface for the two sides to communicate and transact directly

Where to build

Basically any thing with lots of phone calls/emails, manual clerical work, and document transfer:

• Health insurance - how to build a change health competitor

• Brokerage-based B2B categories - how to replicate Faire in other verticals

• Internal communications across large enterprises - how to build a new benefits marketplace

The historical parallel here is Doordash signing up unwitting restaurants, flooding them with inbound and clunky (human) agents, and then selling them Doordash to manage the demand.

How to invest

You either need really big categories ready to adopt such that you can make up for the low margins with tons of volume fairly easily and quickly OR a straight line/short path to flipping the card and selling to both sides of the market without massive volume first.