In 1978, two NASA scientists predicted that as launches got cheaper, orbit would fill with debris until collisions bred more collisions and the lane got hard to fly at all. AI just did this to proposals. The cost of launching one fell to near zero, the lane filled with lookalike volume, and now everyone — bidders, evaluators, the agencies in between — pays the navigation tax.
Donald Kessler and Burton Cour-Palais published the model in the Journal of Geophysical Research: every object you put in orbit raises the odds of a collision, every collision produces fragments, and every fragment raises the odds again. Past a certain density the cascade feeds itself, whether or not anyone launches another thing on purpose. The unsettling part of the math was never the junk. It was the feedback loop.
Today ESA models roughly 54,000 objects larger than 10 cm in orbit. About 9,300 of them are working payloads. Everything else is shells, fragments, and dead hardware moving at closing speeds where you can’t inspect what’s coming — you can only track it or get hit by it. Most of what’s in orbit isn’t a satellite. It’s the debris of one.
Swap launch cost for drafting cost and the model maps one to one.
Note what the metaphor does notsay. Orbit still has working satellites in it, and the debris field is full of genuinely useful AI — narrow tasks, checked outputs, real leverage. The problem isn’t that AI is junk. It’s that junk and signal share a lane and arrive at the same speed, and the buyer’s instruments can’t separate them cheaply anymore. That’s what makes the whole market harder to navigate, for the careful and the careless alike.
NSF has formally told its research community how generative AI may and may not touch a submission. GSA has floated a contract clause aimed squarely at undisclosed AI use in performance. Evaluation panels talk openly about lookalike proposals. None of this is anti-AI. It’s what a range-safety officer does when the lane gets crowded: add tracking requirements. Expect more of it, and expect the bidders who can show provenance — who wrote what, checked by whom, claimed on what evidence — to clear the gates the volume players bounce off.
Operators who keep satellites alive in a crowded orbit do three things, and all three translate.
Know the provenance of every claim in your response: who made it, what backs it, where the proof lives. If you can’t trace it, you can’t defend it — to a red team or a contracting officer.
The cascade is fed by bids that shouldn’t exist. A real bid/no-bid gate is debris mitigation: fewer, better launches beat volume every time the lane is crowded.
Every AI-drafted section gets a named human owner and a check before it flies — that’s the PARC gate. An output nobody owns is debris by definition, whatever model made it.
Before your next AI-drafted section ships, read it against one question: could the evaluator tell it from the other twenty in the stack? If the answer is no, it isn’t a proposal yet. It’s debris with your name on it.
Ten client seats, every bid owned by a named senior operator, every claim traceable. Fewer, better launches — and we’ll tell you when not to launch at all.
Book a call →