Bittensor TAO

Bittensor is an attempt to build a market for machine intelligence. The network is divided into subnets, each defined around a task — producing text, generating images, scraping data — where participants submit outputs, other participants score their quality, and TAO is emitted to whoever the scoring rewards. Its monetary design deliberately mirrors Bitcoin's: a twenty-one million cap and periodic halvings.

The premise is that if you can measure the usefulness of a model's output, you can pay for it without any central employer deciding what is good, and competition will push quality up the way mining difficulty pushes hashpower up.

Measuring usefulness is exactly the hard part, and it is where the honest scepticism sits. Scoring mechanisms have been gamed, restructured and re-argued repeatedly, and subnet economics have been reworked by governance more than once. Most of the activity on the network is driven by the emissions themselves rather than by end users paying for the models, which makes it difficult to distinguish a functioning market from a well-designed subsidy. "Decentralised AI" is a category claim considerably ahead of demonstrated production usage, and the frontier models it is implicitly compared against are not built this way.