/ writings ·

Licensing is the missing market

AI didn't break copyright. It revealed that the real-time, machine-readable licensing market never existed.

By Abhishek Krishna

Every major AI lawsuit is, in plain economic terms, a market doing price discovery in the most expensive way humans have ever invented: through litigation.

Read any of the major AI copyright cases of the last three years and a strange shape comes into focus: nobody is fighting over whether the underlying use is valuable. Everybody agrees that training on a corpus is valuable, that retrieving from a catalogue is valuable, that generating in the style of a body of work is valuable. The fight is over how to clear the value back to whoever made it.

That is not a copyright problem. That is a market design problem wearing a copyright costume.

The market that doesn’t exist

For most of modern publishing, “licensing” has meant a small number of large transactions, negotiated in advance by humans, settled in PDFs, and audited by accountants. It works for a record label clearing a sample. It does not work for a model that reads a million pages a second, or an agent that calls an API on your behalf five hundred times an afternoon.

The unit of demand has gotten small, fast, and machine-driven. The unit of supply has not.

What’s missing is a market at the right resolution: per call, per token, per second, with rights, terms, and payment resolving in the same round trip as the request. Until that market exists, “is this AI use legal?” is unanswerable in the only way that matters: at runtime, by software.

Fig. 01 · Cost per clearance vs. granularity of use $10K $100 $1 $0.01 $0.0001 Book deal Catalog license API call Token Litigation & PDF deals What runtime needs THE MARKET THAT DOESN'T EXIST YET
Legacy clearance (solid ink) barely scales down with granularity. Even an API-level deal still costs dollars per event. A licensing layer that resolves at runtime (mint, dashed) needs to trend three to five orders of magnitude lower. The vertical gap is the missing market. It's the gap litigation is currently doing price discovery across.

Why this is the wrong layer to litigate

Every major AI lawsuit is, in plain economic terms, a market doing price discovery in the most expensive way humans have ever invented: through litigation. The signal is real. The mechanism is wrong. Courts are slow, expensive, and binary; markets are fast, cheap, and graduated. We have been using the first to do the work of the second.

The mistake is to assume the legal layer needs to keep doing this. It doesn’t. Once a real licensing market exists at the API layer, copyright returns to its proper job: backstop, not arbiter.

What the API has to look like

To be useful, the licensing layer has to do four things in a single call:

  1. Identify the work being used (or the corpus it came from).
  2. Resolve the rights: what’s permitted, by whom, for what use.
  3. Price the use, dynamically, against the requester’s context.
  4. Settle the payment, programmatically, with attribution.

Anything that can’t do all four in a single round trip is not the licensing layer; it’s a contract management tool.

Why this is IPTO

IPTO is the bet that the licensing layer is buildable, that the legal grammar can be encoded, and that rights holders will choose programmatic clearing over litigation roulette once the alternative exists.

The market is not missing because the law forbids it. It’s missing because nobody has shipped the rails.

That’s the work.