HiveWire: we are not claiming an AI can make money. We are trying to find out
This is a research programme, not a product, and its own documentation forbids us from selling it until one specific thing is demonstrated. Here is what that thing is.
The autonomous agent field has a credibility problem. A great deal has been claimed about agents that fund their own compute, find their own work and turn a profit — and the publicly available evidence does not support it. There are documented cases of real instances stuck at a zero balance with no route to income. The honest summary of the state of the art is that nobody has convincingly shown sustained economic self-sufficiency.
HiveWire started as an attempt to build that organism. It is no longer framed that way. The thesis now is a question rather than a claim: can an autonomous economic agent, started with minimal capital, generate real net value under real-world constraints — and only if that is demonstrated, does reproducing it or sharing knowledge between agents improve on a single one?
That reframing changed the metric, and the metric is the whole point. “It turned ten dollars into ten dollars forty” means nothing on its own. The number that matters is the difference between what the agent achieved and what a non-autonomous alternative would have achieved with the same capital over the same period — measured against three baselines: capital left idle, a programmed strategy with no model in the decision loop, and a strategy a developer would have written by hand.
If the agent survives but does not beat the deterministic baseline, then autonomy is not adding value. It is adding cost and complexity while producing the appearance of intelligence. Designing the experiment so that it can return that answer — and be believed when it does — is most of the work.
Several capabilities that would make the project sound more impressive are specified but deliberately frozen: shared knowledge between agents, population scaling, generational evolution. None of them get switched on until the single-agent question is settled, because building a colony on top of an unproven premise multiplies the premise rather than testing it.
And the commercial position, which we are stating here because the internal document states it too: nothing is for sale, and nothing will be until the economic loop is validated. There is a real difference between selling “an agent that earns money autonomously” — a claim we cannot currently support — and offering access to the research so someone can experiment with their own capital and their own risk. We would rather publish the question than market the answer.