My AI Thesis: Luck Wins

It's hard to separate skill from luck — that applies to sports, business, investing, and so much more. Often, the role of luck is overlooked, especially in complex systems.

What's fascinating is how luck influences outcomes when baseline skill is equal — otherwise known as the "paradox of skill."

The theory goes: as agents within a field become more skilled, the difference in their performance diminishes and reverts towards the mean, leading to a tighter distribution of outcomes. Said differently, the bell curve shrinks, it becomes harder to outperform, and the role of luck becomes more pronounced in determining who breaks away from the pack (and pulls up the middle and down the sides of said distribution).

This phenomenon is well-illustrated by Ted Williams' 1941 .406 batting average. The league average at the time was around .285, and Ted's record has not been touched since. In fact, the closest anyone has come is .388.

As Michael Mauboussin outlines in The Success Equation, this batting average record—and subsequent compression—is a result of both MLB pitchers and hitters improving. They've collectively become stronger and faster and have better training mechanisms. Inputs and processes have improved, and players have all become more skilled.

Consequently, the collective league batting average has decreased, and the standard deviation has narrowed, leading to more tightly bound performance metrics. While skill does play a role in success and failure, and more skill gives people/products an edge in attaining success, that edge offers no assurance of who comes out on top. Luck does (e.g. the 2019 Washington Nationals).

Now, let’s talk about this in AI and software terms.

Chris Paik recently wrote that Software is Dead. I don’t agree with all of his thesis, but it provides a helpful baseline of assumptions for this piece. Let’s assume (as Chris writes) that AI is as permeable as people believe and anticipate. Software will be easier to write (AI will write it) and product experiences will improve. AI will homogenize software development, leading to lower barriers to entry, increased competition, deteriorating value propositions and profits margins, etc.

But if we apply the paradox of skill, as AI elevates the baseline across the software industry, products—and companies—will revert to the mean.

Side note: while luck and skill are on a continuum, we can define "skill" as a task where cause and effect is evident. You can repeat the behavior and get the same result. For example, manufacturing or software engineering are on the skill side of the continuum because there is clear cause and effect (e.g., a part makes it on the machine and it works, or a line of code is written, and the software does what the code tells it to). In contrast, tasks that involve trial and error with outcomes that are a circumstance of chance/non-repeatable conditions fall more on the "luck" side of the continuum. For example, developing a campaign that goes viral (and, perhaps, “strategy” in general). I'm not saying there aren't people who are more skilled at marketing or creating highly shareable content, but luck often determines more of what will succeed and fail in these tasks than in, say,?engineering.?

That's all to say: If you believe in the assumption that AI will begin to write more code, building generationally great software companies will require more luck – now and in the future.?It will also require more luck to invest in these outliers.

In addition, it also requires more luck to invest in these outliers.

So, if you're a "traditional" early-stage software investor who believes in the platform shift of AI (me), I think you have three options:

1. Acknowledge luck plays a role. Accept that generating outsized returns in application software will become more challenging with traditional investment models. This also means that portfolio construction matters even more. Your ability to back the truck into your winners is more critical when big winners are even more rare.
2. Invest in unique excellence. This is something all of us should be doing already. But it feels more important than ever to find companies/people who are distinctly great at some specific thing?and invest time, money, and resources into that specific thing to create a competitive advantage.
3. (2a...not 3...LinkedIn formatting is weird). To me, excellence in distribution feels like that thing. One strategy is to try to identify which products will gain social influence faster than the competition. Social influence is what explains what happens when a product receives positive feedback – making the strong stronger and the weak weaker. It is a compounding advantage. Often, who benefits from this process of amplification is a result of luck. I believe the biggest beneficiaries of social influence will capture this advantage through influential individuals + groups who are linked to—and refer—their products (e.g., repeat/high-profile CEOs/operators, actual creators/influencers, policymakers, etc.) who wield both influence and distribution power. (Yes, this is a bit of an oversimplification as other factors like timing and consumer behavior also contribute to social influence). However, once an innovation reaches a certain level of popularity, its success is virtually assured. A company's durable competitive edge ultimately comes from its people, culture, and DNA, but it's initial advantage that can tip the scale pretty quickly can be generated through distribution (cough cough, Superfiliate).
4. Invest in weirder shit. Explore the unconventional. Swim where others aren't swimming. To generate alpha, you need to find skill where others do not. This might (and likely will) mean non-traditional product and business models (i.e., not just application software).

In reasonably stable environments, progress requires focus, effort, consistency, and some level of skill. People/organizations can develop processes that improve their skills and inputs, in efforts to generate predictable and replicable success. But I am not talking about progress here, I am talking about outperformance.

I know that the paradox of skill might not apply uniformly across industries and sectors. That said, by acknowledging and adapting to a new AI-driven paradox of skill in software, investors can be better equipped to navigate the next decade of investing — and can hopefully lean on more than luck to generate returns.