Berg's Words


Software and the people that make it

Just about every disagreement that I’ve had among intelligent professionals comes down to a question of time horizon. Should we invest in A or B? Should we take on technical debt or not? Should we build a set of general and composable building blocks to reassemble into a solution for the problem at hand or just plug away on the surface-level problem? Should we pay up for fantastic talent or settle for more mediocre hands? Should we establish prescriptive processes and rigid, top-down edicts about how and which work gets done or should we create a culture of trust and imbue our people at all levels with the autonomy to explore, build, and prioritize? Well, what’s the time horizon on which we are evaluating our return?

I argue that there is only one reasonable answer to the question of time horizon and I have yet to meet many people whose actions indicate that they agree: infinity.

I write this in the midst of a pandemic and the largest disruption to economic activity in my lifetime. I write this as whole industries have disappeared and, rationalize and hope as some might, I remain unconvinced that all will return or look particularly recognizable if they do. I write this as my employer apparently considers layoffs and many millions have lost a paycheck as companies struggle to survive. Still, I believe, infinity is the right timeframe on which to evaluate decisions.

Obviously, on a time horizon of infinity, only a fool would ignore lower bounds and only a charlatan would pretend that liquidity doesn’t sometimes determine solvency. The world is fat tailed and survival comes first, always (err… second. Some things are worth your company dying for). If there is no tomorrow, value at infinity = 0.

So too, fleeting opportunity. Nearly always, the best course is to react slowly or not at all, think, and move intentionally along the path of an infinity-optimal strategy. It requires more thorough treatment than I give it now, but after reading Kaufman’s Origin of Order, I believe there is significant value in large islands of a complex system remaining frozen, permitting percolating change to cascade around them. I would like to one day take the organizational and system dependency structure of real companies and apply the same analyses Kaufman applies to his models and—I imagine—derive quite similar results.

But I digress.

Fleeting opportunity. There come moments when the zeitgeist explodes and the perspective of nearly the whole of humanity converges in turmoil. Those moments come at most a few times (we hope) in any one life. Yet these moments will define the next decades of the human experience and, consequently, commercial winners and losers. To sit back idly at a time like that is lunacy. The smart business arrived at this crossroads with an innate understanding (preferably an articulation) of some crucial aspect of the human condition (true at least for their customer base) and a means of filling a need in consonance with that truth about how people behave, what they desire, or how they interact. That smart business now endeavors to guess at the possible future paths out from the present crisis, determine how or whether their base understanding of their customer permanently changes, and begins placing bets to ensure victory in each such case. Disruption is a breeding ground for convex, outsized outcomes. Though the future may be uncertain, inaction is itself a choice and rarely a good one in the midst of turmoil. Even if none of the core assumptions upon which a business is predicated have changed, there are almost certainly opportunities to rely on the existing assumptions about how people behave to further ingrain the company or its products into the lives of customers and, so, to improve the infinity-optimal outcome.

Yet nearly all of a businesses existence is not spent eye-to-eye with the lower bound of a bank balance or hand-to-hand grappling with destruction and disruption. The vast majority of the existence of most companies, therefore, should be spent ignoring much of the day-to-day, shutting out the weekly advance or decline of the competition, and focusing on its strategy for winning at t=inf. A successful strategy has to start from a true belief about humans or a subset thereof and a means for getting them something they desire. Those basic inputs cannot possibly change day-to-day or month-to-month so any change in output is merely a reaction to noise or an admission that the basic requirements for a strategy never existed in the first place.

Compounding is the route to riches. That means that anything you can’t repeat consistently doesn’t matter except insofar as some one-time move increases the time you have to compound (keeps you alive) or changes your long-term growth rate.

But why infinity? Why not just some other long time horizon over which gains can compound? Because human interaction does not exhibit optimal substructure and, therefore, greedy algorithms are largely inapplicable. On the contrary, our world exhibits extreme path dependence. Reputation and brand—the subconscious response the representative consumer has to a seller—reign supreme. There is no reason to believe that the configuration that optimizes total profit at t=T is a point on the path to the configuration that optimizes total profit at t=T+ε. Any finite timeframe yields the same problem. A metrics-driven, monotonic march likely doesn’t asymptote to value-maximization.

One subsumed but important sub-point of a lack of optimal substructure in complex human societies: a time horizon of infinity breaks the backward induction that leads to the most insidious principle/agent problems afflicting corporations. I can very honestly tell an engineer working in an infinity-optimizing company that the route to their greatest personal success is to simply focus on growth rate—spend as much time as possible just focused on becoming a better and better engineer. Ignore promotion timelines—they’ll come when they come. Ignore the insults of this year’s comp or last year’s bonus—remuneration over the long run will more than compensate. Ignore the unfairness of that lesser engineer in the other org who got promoted faster—competence shines paramount over time. The best route to personal success in an infinity-optimizing company is maximizing value produced for the company over time, which is typically best accomplished by self-improvement.

Not so in a finite-horizon company. On finite time scales, pies don’t grow much and the size of the slice you get today matters a hell of a lot to your lifetime success. On six month planning cycles, that promotion you missed last half might well be existential. That stretch project that could pay off big for the company and for personal growth might not make quite so much sense.

I never want to care about when I get promoted. I never want to ask for a raise. I would like to work at a place that creates the preconditions for aligned incentives—namely, an infinite time horizon. In the short run, you have to care about your slice of pie. In the long run, you can rationally focus all of your effort on growing the overall pie. Growing pies is just more fun.