Monkey See Monkey Do

_AI is an imitation machine._

May 21, 2025

• If something’s visible, it can be imitated.
• If something’s visible with invisible components, it can be approximated.
• If something’s invisible with visible components, it’s a bit of a mystery.
• If something’s invisible, it’s a total mystery.

The imitation game applies to all kinds of media: artwork, music, writing, code.

Before AI, we were in a Newtonian age. You could observe a production system and predict with reasonable accuracy how it would evolve over time based on the company’s roadmaps and release notes.

After AI, we have entered a period of high uncertainty and systemic instability.

Heisenberg’s Uncertainty Principle comes to mind: The more we know about a particle’s velocity, the less we know about its position, and vice versa. It’s not a perfect analogy, but I’ll propose the following:

1. If you observe an organization with high velocity (i.e. top developers equipped with the latest AI coding assistants), you cannot know their current position (i.e. the state of their development efforts) with any certainty. They could be bogged down with unpredictable interactions between AI models and their legacy code base, or they could be making giant leaps in directions you cannot anticipate.
2. Even when you do observe an organization’s position (i.e. the state of their production applications), you’re still in no position to accurately assess their velocity (i.e. AI coding capabilities). Something that went into production today may have been created with tools released six months ago. Something that goes into production next month may have been created with tools released yesterday. The pace of delivery is getting faster and faster. We’re getting to the point (in software) of: “If you can describe it, AI can build it.”
3. If an organization’s velocity is zero (i.e. they’re not using AI), their position will remain fixed, predictable, and ultimately indefensible. And because nobody wants to be perceived as a zero-velocity organization, you’ll see companies claiming to be more advanced than they are. That posture cannot last long.
4. High-velocity legacy software companies are among the first to realize how thoroughly AI will disrupt their existing business models. They’ll do their best to control the pace of change – not too fast, not too slow. They’ll also try to influence the narrative around AI and tilt the regulatory environment in their favor by currying favor with even the dodgiest of governments.
5. The biggest uncertainty comes from high-velocity unknowns without legacy business models. We saw an example of this with DeepSeek, which thoroughly upended the narrative that creating AI models was a capital-intensive business with economies of scale. I’m confident that we’re going to see more breakthroughs with highly disruptive outcomes coming from unexpected places.
6. As the software development lifecycle shortens to T+0, the cost of software (including enterprise software) tends towards free.

And what happens when creators, makers, and civic organizations cut out software licensing fees? That opens exciting possibilities for everyone.

And that’s why I’m an AI Accelerationist.

Let’s make more independent breakthroughs.

The sooner the better.