Directly customer facing is a tough one, since a lot of the time chatbots are used to just fob off customers that eat up a lot of time (old people complaining etc).

I think the best experience I've had is when companies offer an MCP server that lets me get up and going with their product very quickly. Sequenzy is one such example I tried, very AI-first and works well with Codex.

If you're talking about stuff that isn't customer facing, there's all kinds of tasks that can and have been automated.

As a simple example, I run a retro games business and we've vibecoded up various controller/cartridge testing apps that have dropped the time it takes to test a box of product by around 50%. I'm planning to spin up a few Hermes agents soon to watch the prices of things and suggest improvements to our listings (e.g. noticing we haven't soldas many copies of Super Mario 64 as we should, and checking our keywords/pricing).

Obviously puts in big work with SaaS and stuff too.

EDIT: actually, Shopify's built-in agent works really well and is directly customer facing. It's a smart model (I think either a Kimi or Qwen as a base) that seems to have been fine tuned on their product itself.

You can ask it stuff like "what products have seen a decrease in sales" or "what is the value of my next payout" and it just answers them.

Before you had to manually write DB queries for this stuff, or more realistically just ignore it.

MCP is just API, and we have been there already and we know what happens: Businesses realise they can’t extract maximal value if they don’t own the interface so they recant the API.