A quick recipe for LLM-based products that deliver long-term value

One of the easiest ways to monetize LLMs is to sell prompts, or to wrap prompts in a product. For example, "download 1000 marketing prompts" or "generate your GTM in 3 easy steps". There's nothing wrong with this approach. It's a good time to squeeze these side-hustles now that the world of LLMs is so new.

But if you rely solely on the quality of your prompts and your prompt chains (and maybe your UX), it's likely that your early adopters will not pay because they probably know how to do it in ChatGPT themselves. And even if laggards are willing to pay, then be sure you'll soon be in a red ocean because prompts are easy to replicate.

IMO, to build an LLM-based product that can deliver long-term value, here are some key ingredients:

1. Use of an LLM (obviously)
2. Use of functionality outside of LLMs: Incorporating external functionalities like APIs or databases gives the product specialized skills that a standalone LLM cannot offer, adding unique value. You just have to build or find services that deliver incredible value when mixed with an LLM.
3. Automation of work: The automation should target tasks that are sufficiently time-consuming or complex, ensuring that users see the product as more than just a wrapper around an LLM. It should be so time-consuming to do the same thing in ChatGPT by copy-pasting prompts, that people would prefer to pay for your service. This elevates the service from something they could "just do themselves" on ChatGPT.
4. Great prompt chains: A sequence of well-crafted prompts not only guide the LLM to offer nuanced and contextually relevant responses, but also can cover edge-cases very well. This will require trial and error, and you have to make sure you have an automated QA system to constantly check your prompts, as models update.

A quick rule of thumb to see if your LLM-based product can deliver long-term value:

How much more value is your product delivering, than a blog post which details your prompt chains? If the answer is less than 10x, you probably will have issues monetizing it long-term.

Now, these guidelines are unique to LLM-based products. But standard advice for any product still applies here: niche down, leverage network effects, partnerships, etc.

Another approach, which I'll cover in another post, is to build products that become the foundational technologies of LLMs in the future (e.g. Pinecone).

Last point I can't resist the urge to mention:

Prompts are seeds, not gems.

Don't treat your prompts as proprietary secrets. This field has so much to offer and we haven't even seen the start of it yet. Share them to foster innovation.