A world of uncertainty

I wanted to share some ramblings about some of my experiences lately.

As CTO of my last company, in a world where AI was still not good enough to meaningfully accelerate development, I was used to systems being built and maintained at the speed of human cognition. Which is to say insanely slow by today's standards.

The planning horizon for software projects in the past was much longer. You would carve up some work, plan the architecture a few months out, and then the project would unfold at a semi-predictable pace. And there was a sense of comfort and control in that. A machine, totally obedient, fundamentally understandable in its essence, subject to clear laws.

I have to say, I miss that sense of control. The sand seems to be shifting beneath our feet every day, and the operating environment is so unstable that we can't be sure what even the world will look like six months from now. Anthropic might become the god ASI that nationalizes and merges into the government, but we may also arrive at sovereign, decentralized intelligence before they can do it.

The mentality that I brought to my projects in the past is fundamentally unsuited to our world today. Control by planning, the idea of a stable period of execution of a well-designed plan — all of it has the markings of the failures of the generals of the early 20th century, with their elaborate timetables and plans, the commencement of an operation that once begun could not be stopped. The war could not be stopped because their plans could not account for the infinite variety of branches that might unfold once the first domino had been tipped over. The cone of uncertainty is now simply too great.

I have found myself needing to evolve new decision frameworks as a result.

Instead of elaborate plans of long sequences, I now think in terms of immediate uncertainties, and the minimally costly tests that would resolve them. It is preserving strategic ambiguity; it is maintaining optionality branches where information is insufficient to distinguish a path's expected value. It is basically a Bayesian model of the world, an attempt to be as serious as possible about the epistemological limits of our mental models. And most certainly a drastic departure from the world of comforting predictability that I once knew.

But I am letting go. And in that, one can find — as Bruce Lee says — a flow like water.