We’re building the infrastructure for a future AI co-scientist, starting with computational research domains. Our thesis is that capable AI scientists require two things built in tandem: (1) a human-centric product suite that generates high-quality process data, and (2) the research infrastructure to turn that data into increasingly autonomous systems. We’re creating the product and training infrastructure needed for research agents that can execute the full scientific loop. Synthetic Sciences Web is our platform where scientists delegate complex research tasks to swarms of AI co-scientists across reading → hypotheses → methods → experimentation → results → drafting, even while they sleep. In parallel, we develop RL environments and process-based training data for LLMs, starting with agentic coding environments for ML research workflows.