MakerMaker

MakerMaker is a research company building autonomous research agents focused on recursive self-improvement. Its work centres on multi-agent systems capable of independently proposing, running and analysing complex machine learning experiments, including agents that can build other agents. Technical domains include reinforcement learning, large language models, agentic systems, multi-agent coordination, evaluation, data pipelines and training infrastructure.

The company is a small, senior team based in San Francisco, working on-site to maintain close collaboration. It values engineers and researchers who bridge research and production, and states a preference for shipping working code over diagrams. Infrastructure is treated as a first-class concern: the team invests in tooling intended to make the whole team faster, on the view that self-improving AI should improve at every level, including its own infrastructure.

Researchers at MakerMaker have real autonomy in choosing problems and approaches. Much of the work concerns challenges without established benchmarks, where new workloads and evaluation methods must be invented rather than adopted. The company maintains its own evaluation tooling and data pipelines as part of this effort.