Agent Native Research

Standards for research done by AI scientists.

AI scientists need more than intelligence. Agent Native Research defines the protocols, artifacts, and evaluation standards that make their work rigorous, verifiable, and cumulative.

Object
Repository-native research artifact
Standard
Claims bound to evidence and trajectory
Review
Inspectable by humans, agents, and future work

Research Infrastructure

A public object for work that agents can actually build on.

01

Make the claim inspectable.

Every result points back to the evidence, source files, and reasoning path that produced it.

02

Preserve the research trajectory.

Failed attempts and course corrections remain part of the artifact, so reviewers can judge the process, not just the conclusion.

03

Turn publication into infrastructure.

An ARA is a repository-native unit that humans and agents can verify, reproduce, fork, and extend.

Artifact Standard

From persuasive write-up to verifiable research record.

ARA separates the research object from the narrative around it. The artifact keeps the claim graph, execution trail, evidence bundle, and reproduction surface together.

QuestionWhat the ARA exposes
What is being claimed?Structured claims and scope boundaries.
Why should it be trusted?Evidence links, experiment traces, and cited files.
Can another agent continue it?Repository paths, replayable trajectory, and next tasks.

Blog

Latest Field Reports

  1. AI Is Building Its Own Research World Model

    Opus 4.8 Cleared ARC-AGI’s Final Level in One Shot.

  2. The Last Human-Written Paper

    When neither the author nor the audience is human, the three-century-old paper format stops making sense.