I’m writing this in mid-2024. GPT-4, Claude 3.5-Sonnet, and Google Gemini 1.5 Pro are the world’s leading LLM models, and the ever rumored Llama-3-400B is supposedly just around the corner.

The LLMs are wildly capable: they don’t just write sonnets, or teach you about history, but they’ve even turned into amazing coding assistants. Sure, they still hallucinate sometimes, but most of the time it’s not worse than your average person (or stack overflow answer).

The problem is that no one knows what the future is going to bring–will RAG rule the day? will we need new models with flexible depth-of-compute as per Yann LeCun? will the scaling laws finally fizzle out? or will we simply run out of voltage transformers?–just that it’s going to happen soon. And with tens, if not hundreds, of billions of dollars in play, the mega AI labs feel like behemoths, ready to wipe out entire business models in a glance, or at least a model update.

So what do you do if you still want to play in this space?

The way I see it, there are three main options: find a defensible niche, adapt at the edge, or build foundations.

Find a defensible niche

There are some industries that are hard to enter, or might be too small to be worth it for the big players, or require some kind of specialized knowledge, infrastructure, or process to tackle. Think industries like healthcare (especially with HIPAA), defense tech (clearances, govt contracting), or autonomy (specialized training datasets).

The fundamental bet here is that the extra barriers to entry will give you enough headroom to establish a sustainable business.

Adapt at the edge

If you’ve got the network for it (or the dedicated manpower, or connections, or pre-existing business lines), it might be possible to build a reactive business. That is, to always be on the look out for “the next step” and to simply build out the new products faster than your competition (once it’s clear what they should be!).

This is a road of constant exploration and vigilence. It means you’ll need to do a lot of networking, outreach, and learning, and probably also have your own internal AI R&D effort, even if only for the second-order benefits of improved market insight. You’ll probably also end up sponsoring events and truly trying to build a community. After all, you won’t know precisely where the next product signal will come from… only that you’ll have to find and act on it faster than your competitors.

Build foundations

There are a lot of people within the AI space. I just got back from Seattle, where CVPR broke 12,000 attendees. And yet, that’s a small fraction of the population. For everyone right now pushing the edge of AI systems, there are a thousand more who know they should care, who know they’ll need it, and who wish it was within their reach, but who simply don’t know how to get started. How to put the pieces together and make things happen.

Building foundational tools–note! not foundational models!–for these people will be a great way to bring value and earn happy customers. But what counts as a foundational tool?

The operative question is, for any particular customer segment,

“what can you build that will embody the lessons-learned from the experts-at-the-edge, so that your customer can build their own systems successfully, and without needing to pay the cost to learn those same lessons first-hand?”

My favorite example here is not a software product at all, but the bodies of documents we call building codes. Many lessons have been learned over the years about fire safety, earthquake sturdiness, handicap accessibility, and more, and the beauty of building codes is that they oblivate the need for individual builders to learn those impossibly-many items for themselves. Instead, they establish a standard framework of “18 inches on-center”s and “standard weight Schedule 40 wrought iron”s that empower the end users to put the pieces together for themselves faster, and more confidently and easily, than would otherwise be possible.

Of course, this works for software just as well. “Build the systems that let people build systems” is a time-honored strategy.