Right now I bet there are dozens of startups trying to figure out how they can build AI assistants to be even better intermediaries. The business plan looks something like:

• Find some area where you can put in a bit of work and get an AI assistant that is seriously useful to some group of people. So useful that they love it and get their friends to start using it. Maybe you start with people trying to buy houses who get a ton of mostly useless emails with new listings but also the occasional gem. It doesn't matter whether there's a commercial tie-in at this stage, though, you just want to be great.

• Keep expanding the areas your AI assistant is useful in. Talk to your most enthusiastic users, find related areas you can save time and hassle, grow, grow, grow.

• Some of these areas will have natural commercial applications; make deals where you send them a bunch of traffic in exchange for a small discount, and keep part of the discount. Disclose this to your users so the FTC doesn't get mad at you!

• Stay focused and execute better than your competitors. It's a crowded space, but also one with a lot of potential.

• Eventually you have a large number of people where you're buying things on their behalf and exercising substantial discretion in doing so. They're happy with your purchases, and your position as a trusted intermediary lets you bring in a ton of money.

I haven't yet met an AI assistant I'd trust to act on my behalf, and prompt injection is pretty concerning. But as companies figure out scalable oversight and improve reliability I expect we'll get there.

(Prompted by helping a friend think through a business plan for a startup, though with how crowded the space is they weren't able to get funding and decided to do something else.)