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?
You stretched the analogy too far, to the point it doesn’t work at all. The advice to sell shovels works because shovels are cheap, already invented, simple, proven to work for their purpose, and can be bought in bulk. None of that is true in your analogy.
If anything you have it backwards and the LLMs (or eventually AGI) are the “gold” everyone is chasing. The low-effort crummy products that shove AI for the sake of hype and parting fools (mainly greedy investors afflicted by FOMO) from their money are the “pyrite”. A shovel in this sense would be something OpenAI, Anthropic, Google, DeepSeek, etc would want.
The shovel analogy is typically used to discuss tools versus goals, irrespective of the microeconomics of buying and selling.
I can use another sector, biotech, as an example. The sequencing industry spends billions of dollars designing and selling the next generation of sequencers, along with collecting more and more data. They are still actively researching and developing new platforms. The end goal (product) of biotech is ultimately to produce a medicine. Pharma is in the business of end products. You can spend billions on sequencing and never produce a new medicine.
Shovels = sequencing
End product = medicine
That has nothing to do with the cost or availability of shovels. As a biology major, you could choose to go into sequencing or go into medical research.
AI is a tool. For what? In the future, who knows--isn't that the big question at the moment, about whether the investments are justified and so on. But other businesses will presumably purchase and use these AI services. OpenAI can spend $500 billion, but how are they going to recover the cost (short of AGI demolishing HFT or the like, or simply taking over the world) without another business buying it and using it to create value in the form of end products.
Shovels are typically B2B.
End products are B2C.
OpenAI doesn't need anything from (the proverbial) you. Also, I stated in the title, "Currently". We aren't talking about doomsday AGI situations where AI exists as the end-all of humanity, and the point is just to own it. For current investors and labor, the model appears to be that these large companies sell their AI services.
I would argue at least 3 of the 4 you mentioned are the shovels in AI. They are what everyone uses to build AI tools, which is the gold in this analogy. Anyone that offers infrastructure - hardware or software - around building AI is selling shovels IMO. Of course, to complicate things, they all sell applications to the enterprise too, so they’re not pure shovels.
The other 3rd category that are shovels is anything that provides data to LLMs. This include media publishers like Reddit. They definitely are not hard tech.
Anyone that is building any tools or applications whose target market is not companies who are building AI applications, is chasing after gold.
I think in this analogy the shovels could be blogs, podcasts, youtube or other content produced quickly that offer info about $AI_thing. This is all mostly just marketing for the person’s content platform which is monetized and is often low quality content that is a regurgitation of press releases or release notes, etc.
Your analogy makes no sense and you took it too far.
But trying to grok what you are getting at. One of my consulting focuses is on end user products and integrating LLMs into call centers - both voice and text. Not the dumb - “create a prompt and let it run wild”. It’s much more controlled. This is a really simplified example.
https://chatgpt.com/share/678bab08-f3a0-8010-82e0-32cff9c0b4...
You can imagine once you have well formed JSON, you can treat it like any other request on the backend.
None of those,
First, you build something that fixes a customer problem that they are willing to pay for.
Whether it uses AI or not, the customer doesn't care.
The only thing that matters is the value you add and your ability to sell and showcase how your solution is better and attract customers.
The answer is yes.