AI agents are solving unsolved problems, and Bitcoin is evaluating them—is it too early, or is it just hype?

The dreamwalk project on GitHub aims to combine two of the hottest trends at once: autonomous AI agents and cryptoeconomics. The idea is that agents don’t just generate ideas; they go through a “living research marketplace”—complete with a blind jury, real-world experiments, rules, and a graveyard of failed hypotheses. Evaluation is conducted in Bitcoin.

The repository contains a set of files that resembles an experiment’s constitution more than actual code: ATLAS_CONSTITUTION_v0.txt, PROOF_OF_WALK_SPEC_v0.txt, PROTOCOL.md, HARNESS_MODE.md. The names speak for themselves—this is an attempt to establish formal rules for a process in which AI agents act as participating researchers rather than as tools in human hands.

The key promise is “novel-idea brain is free, plug in.” In other words, the core idea-generation system is open, and anyone can connect their agent. Then the mechanism kicks in: the agent proposes a hypothesis, goes through a “blind court,” and the result is recorded. Failed ideas are sent to the “graveyard”—which is an interesting idea in itself: rather than hiding failures, they are made part of the protocol.

The problem is that, despite the concept’s richness, the repository has zero stars and zero forks with 29 commits. This isn’t a product or a platform—it’s a draft specification that no one has picked up on yet. And the main question isn’t whether the code works (there isn’t really any), but whether the model itself makes sense.

Economic incentives via Bitcoin for AI research—an idea with two fundamental problems. First: who evaluates the evaluators? “Blind judging” sounds good, but if the judges are also language models, we end up with a vicious cycle where an LLM evaluates another LLM, and the cryptographic mechanism merely adds the illusion of objectivity. Second: real scientific experiments require not tokens, but laboratories, data, and reproducibility—things that no on-chain protocol can provide on its own.

That said, this approach shouldn’t be dismissed. If we look past the crypto wrappings, dreamwalk raises a genuine architectural question: how to organize the collaborative work of AI agents on research tasks in a way that ensures a selection mechanism, the documentation of failures, and reproducibility. It is precisely the “graveyard”—a list of dead hypotheses—that may prove to be the most valuable part: a structured archive of what has already been tried and why it didn’t work is just as necessary for an ecosystem of agents as it is for humans.

For now, this is just a concept on paper with an ambitious claim and zero validation. But if someone takes the idea of a protocol for collaborative AI research from here and drops Bitcoin—it could turn into something worth keeping an eye on.

Source: Hacker News - Newest: ""AI" "LLM""