I recently saw the following tweet from Bryan Johnson which encapsulates the emotional impact of the rapid improvement of Large Language Models.
While he jumped to the conclusion that we all should prioritize longevity, this wasn’t a satisfying answer for me at least on an individual level.
Some of us may be left wondering what to do with our careers in a world where our understanding of what is prestigious, fast-growing or impactful is changing faster than we can adapt to it.
In crypto, we’re already frequently forced to reassess our focus and whether it’s creating anything meaningful beyond short-term value generation.
But before we understand how to operate in this new paradigm we have to examine its characteristics.
The bottleneck is evolving
One way to interpret what is happening is to understand that we’ve been living in an era where the rate at which we can produce software is the primary bottleneck.
There are a couple different ways of looking at this:
The most competitive and fast-growing companies are software companies. These also have some of the highest revenue multiples (especially SaaS) due to the attractiveness of the business model.
Sam Corcos, the CEO of Levels even argued that software engineers are so important that start-ups should accommodate for them in every way possible
Many of the most “attractive” careers are built around software: software engineer, designer, product manager, venture capitalist
While software is technically hovering at only 10% of US GDP, it arguably penetrates and affects every other area (yes including services) to an extent that makes categorizing things as software and non-software laughable.
Enter LLMs
What’s special about LLMs is that they contribute to the growth of software in two ways:
Directly by increasing the efficiency of software engineers
Through replacement of software (e.g., making purpose-built software obsolete and replacing it with an AI front-end)
As such they have the potential of massively progressing our current bottleneck potentially to the point where it’s not a bottleneck anymore.
Furthermore while pre-LLM software could be seen as a tool, LLMs are making the separation between labour and technology much more difficult to draw.
What’s next?
The emergent intelligence in LLM construction makes it all but impossible to predict their capabilities at a 5-10 year timeframe so scenario prediction and strategic response, a common way corporate organizations respond to evolving trends, is not viable here.
Instead, I see people exploring several personal policies:
Look for scarcity. While the reach of software is undeniable, software moats relate more to the world software operates in rather than the code itself. Software is not reliably seen as a store of value itself (especially not now). Instead, the best software companies have dominant network or switching cost effects that relate to human limitations as buyers and perceivers of software value. I believe that human attention (distribution) is a clearly scarce resource and one that may remain scarce throughout the first generations of mainstream LLM adoption. The rise of the DevRel role is not accidental and will continue to accelerate as projects realize that their costs of producing software are going down.
Ignore the change and focus on short-term exploitation. Borrowing from machine learning, In any reinforcement learning model, the high-level relationship between rapid change of the objective function and level of action is well understood.
Make bets on which companies will drive the changes. Predicting which software companies will win is becoming increasingly difficult, but this method relies on being a better forecaster. The reward, of course, is greater as the stakes are higher.
After a brief ”AI retreat” in 2023 I’ve been relying on the first approach.
It’s helped me move away from seeing software as inherently valuable and focusing other moats built around human attention (of which network effects are only one example).
How are you dealing with the future both strategically and mentally?