The AI Research Comet

Structure of a comet from ESA

The AI research frontier is a comet screaming through space—a white-hot nucleus of capability burning forward at relativistic speeds. Behind it, stretched across months and years, trails the debris field: first productization, then market adoption, each layer moving slower than the one before it, each still glowing with reflected light from the source but cooling, solidifying, falling further behind.

The Temporal Cone of Innovation

I've been watching this structure emerge with an almost physical clarity. At the very tip—the research labs, the frontier model releases—progress compounds monthly, sometimes weekly. A capability that didn't exist in January becomes infrastructure by March. The paper that breaks ground on Monday is obsolete by Friday. This is the engine, the core, where the laws of normal business time simply don't apply.

Then, at a fixed distance behind: productization. Here, founders and builders attempt to catch the light, to grab a capability mid-flight and wrap it in APIs, interfaces, business models. This layer moves faster than traditional software—there's urgency, there's venture capital, there's the drumbeat of "move fast"—but it cannot move at research speed. It takes months to build what the labs discover in weeks. The lag is structural, not a failure of execution.

Further back still: market adoption. Enterprises running pilots. Users changing workflows. Cultural acceptance of AI-generated output. This is the slowest material in the wake, the heaviest debris. It moves at the pace of human psychology, institutional inertia, and trust-building. Years, not months.

The Impossible Position at the Front

Building at the research frontier means building on ground that is actively liquefying beneath you. I think about the founders I know who are racing to productize GPT-4-level reasoning, only to wake up to GPT-4.5, then Claude 4, then whatever comes next month. The foundation keeps upgrading. The abstraction layer you built becomes obsolete not because you failed, but because the physics of the problem changed.

There's a particular kind of stress here that doesn't exist further back in the cone. It's not the stress of competition—though that exists. It's the stress of ontological instability. You cannot solidify your position because the ground itself won't stop moving. The major labs aren't competitors you can outmaneuver; they are the terrain itself, reshaping under your feet.

Every time you get close to product-market fit, the product category shifts. Every time you build a moat, the river changes course.

The Arbitrage of Distance

But here's the turn: the lag is not just a liability. It's a temporal arbitrage opportunity.

Those building further back in the cone—six months behind the frontier, a year behind—are building on more stable ground. The capabilities have crystallized. The user expectations have formed. The business models have been tested by the early pioneers. There's friction, yes, but friction is what allows you to gain traction.

The founders building "boring" AI products—tools that automate workflows using models from 2023, wrappers around GPT-3.5-level intelligence—are often building better businesses than those at the bleeding edge. Not because they're smarter, but because they've chosen a different position in the wake. They've traded cutting-edge for compounding momentum.

This is the hidden wisdom of the cone: speed is not evenly distributed, and shouldn't be.

The Existential Choice of Builders

So where do you build? Do you stay at the nucleus, accepting that your work has a half-life measured in months? Do you position yourself in the productization layer, knowing you'll always be catching up but might actually catch something? Or do you wait further back, where the light is dimmer but the ground is solid?

I don't think there's a universal answer. But I do think we've been telling ourselves the wrong story. We've been acting as if everyone should race to the front, as if proximity to the frontier is inherently valuable. But the cone teaches us something else: position is strategy. The question isn't "how close can I get?" but "at what distance can I actually build something that lasts?"

The labs will keep moving. The nucleus will keep burning forward. Your job is not to become the comet—you can't, and you'll burn out trying. Your job is to find the right distance in the wake where your particular mass, your particular momentum, can create something that survives the journey.