Ask HN: Currently, is it better to work on AI shovels or AI applications?

Typically, conventional wisdom has been to sell shovels during a gold rush. Except people aren't quite selling shovels as much as they are still inventing them. You could spend a year designing and manufacturing your shovels, only to have the guy next door come out with the Shovel-o-matic 2000 and render your work obsolete in an instant.

I would consider selling shovels to be anything that can be classified as "hard tech". No, hooking up a LLM to a wrapper doesn't count. That comes next.

I would split up applications into two distinct types: mini-shovels and end products.

As an example of a mini-shovel, consider a shop that sells toys. They have to manage and track their inventory. Someone spins up an AI system to automatically input sales data, update inventory, and place orders for low stock. That saves on staff time and costs. The person who made this AI system goes around selling it, primarily in a B2B fashion. Yes, this "application" helps the toy store a little, but it's not really serving the main purpose of getting people to buy toys.

The end product application is a little more obvious. A toy designer with no manufacturing skills uses AI to prototype complex 3D patterns, iterate, and come up with an ingenious mechanical design. He proceeds to sell that toy directly to customers (B2C) or to toy shops (B2B). The use of AI is irrelevant to the final paying customer.

Of course, it can be more nuanced. The inventory system IS the end product for the software designer. Mini-shovels can become obsolete in an instant, too, but they could have slightly deeper moats depending on industry inertia among an unsophisticated client base. The market scope and dynamics are also very different for B2B versus B2C.

I am seeing a lot of mini-shovel applications, the typical low-effort examples like building a AI-based financial advisor, a new AI-based writer assistant and so on. I assume that it's more difficult to tell how much AI is being used in end products (other than obvious AI art on printed shirts and such), especially if people have no incentive to talk about it.

I'm kind of thinking at the moment that both true shovels and quality end products are difficult. Mini-shovels are easy, and that's why they are a dime a dozen at the moment, but I'm not really seeing their value for creator or clients. I can see that you might be able to make a quick buck, though, if that's your thing.

So, what's your take HN? Shovels, mini-shovels, or end products?