This first appeared on my newsletter Boring Problems, where I write about how fundamental progress is made by tackling unsexy problems that matter. See more and subscribe here.
LLMs are changing change how we build technology and engage with the world. So there are thousands of people trying to figure out how to build and capitalize on this technology wave. But thinking about this top-down when you’re not an expert puts you in the bucket with everyone else, relatively undifferentiated, in a fast-changing landscape.
Suppose you’re not one of the, say, 1000 best-suited people to build core LLMs. Then let me recommend something:
Stop worrying about where LLMs are going, and build something you know. LLMs will find you.
Thinking of this bottom-up from an area of customer need or conventional technology you know keeps you in a tiny bucket of similar people. A considerable amount of value from LLMs will accrue to the domain experts who understand their industry and figure out how to build it for a new generation that includes LLMs.
Tomas Pueyo framed this well in a post called ‘How to become the best in the world.’
Here’s his graphic that best explains this:
If you throw yourself into the bucket of talented generalists trying to participate in the generative AI revolution, you’re probably far from the best. If you stay in your area of specific expertise, you’re already one of the best. If you build there while being at the forefront of applying LLMs to your area, you’re immediately one of the best in the world at that combo.
And a huge amount of value is going to be created and captured by domain, industry, and customer problem experts who start there, and then rapidly evolve LLMs into their software.
A Checklist For Making This Happen
There’s a set of conditions and actions needed to make this work.
You’re solving an actual customer problem. The customer need can’t be ‘use AI.’ It needs to solve a problem where there is or can be budget.
Build where LLMs are some, but not all, of the solution. This won’t work if LLMs have no part of the future solution. They also won’t work if they’re all of the solution (‘chatbot for industry X’). Ideally, you want most of the solution to need conventional, boring technology innovation and some to be served by LLMs. I’m a big fan of areas where LLMs can plug a UI or data gap, unlocking the next generation of experience or power for business users.
Future-proof your architecture. Ensure you’re building with the understanding that you’ll want to integrate LLMs deeply into the product and technology, and this target area is moving fast.
Become an almost expert in LLMs. Or an expert in how they can apply to your area.
Have an organizational radar. Whether that’s you as a leader or a small team that focuses on this, you’ll need to build an organizational radar to integrate LLMs into parts of your product as opportunities present.
Have organizational nimbleness. Speed is important, but not enough. Your team will have to respond and adapt more quickly than we’ve seen in a long time.
Wrap all of that up in a coherent but flexible roadmap.
Stop worrying about how to participate in the generative AI wave and just start where you’re already an expert. Trust that LLMs will come to you, and when they do they’ll help you unseat a generation of tools who weren’t built for them.
This first appeared on my newsletter Boring Problems, where I write about how fundamental progress is made by tackling unsexy problems that matter. See more and subscribe here.
