AI Era Laws of Physics

The two axioms and their reversals

Ben Horowitz argues that two rules treated as axiomatic in software for thirty-plus years have reversed in the AI era, a framing he states as "the laws of physics are different," implying the inversion is structural rather than cyclical.¹

Axiom one: money cannot solve the problem. The old rule held that throwing resources at a software gap did not close it. If a competitor was two years ahead, hiring a thousand engineers would not catch them, the mythical man month problem: nine women cannot have a baby in a month. That was the foundational reason software moats held up under competitive pressure, since capital alone could not close product gaps. In the AI era, the rule reverses. GPUs, good data, and sufficient capital can solve basically anything in software. The constraint is no longer calendar time or team coordination, it is compute and data, so a two-year product lead is not safe if a well-capitalized competitor decides to close it. Competitive moats built on accumulated technical complexity are structurally weaker, because the feature set can now be regenerated from scratch at GPU speed.¹

Axiom two: possession of the customer means multiple lock-ins. The old rule held that having the customer gave a company three independent lock-ins: migration pain from moving data and workflows to a competitor, data lock-in from years of company data living in the system, and UI lock-in from users having learned the interface. In the AI era, Horowitz argues all three are effectively gone. Migration pain shrinks because code is trivially replicable at AI speed. Data lock-in shrinks because clean export formats, APIs, and AI-assisted migration tools have cut switching friction. UI lock-in shrinks because future users will increasingly be AI agents rather than humans navigating a graphical interface: "AIs are really flexible on how they use user interfaces," and an agent does not care whether the UI is familiar, it uses whatever interface the underlying API exposes.¹ Pricing power anchored to these three lock-ins is under structural pressure.

What remains valuable

Horowitz does not argue that nothing has value, only that the value must come from something distinct from the old lock-ins: relational infrastructure such as explicit relationships with global counterparties that take years to build and cannot be bought with GPUs, for example Navan, the travel-software company whose relationships span every airline, hotel, and train line globally; channel lock-in to specific buyer personas that larger models tend to ignore because they are not glamorous; domain-specific data moats built from data AI cannot generate, such as real transaction histories or proprietary sensor data, so long as the data is genuinely irreproducible and not just large; and communication network effects, the form of lock-in most likely to survive, where switching means losing the people a user communicates with.¹

The diagnostic question

Horowitz's frame for a CEO navigating the dislocation is: "Are you getting stronger in that meanwhile, or are you degenerating?"¹ He lays out three failure modes. Customers may simply be buying other things, not yours, which is a deep problem requiring a cut and a pivot. A company may still be generating revenue while its pricing rests on old lock-ins that are eroding, which creates margin pressure until new value is found. Or a company may still be strong while the market has written it off based on its category label, which is not immediately fatal but is only safe if the underlying lock-in reason is still real.

The claim also sits in tension with Taste as Moat: if money can now solve most software problems, the moat lives in judgment, in deciding which problem to solve and how to configure AI against it. The two ideas are complementary rather than contradictory: the old lock-ins are gone, and taste is one of the new sources of advantage. It also connects to No Moat in Software, the parallel argument that any piece of software can be trivially copied once AI coding tools mature.