Maria Gorskikh is co-founder and chief executive of Maritime, an AI-agent hosting platform, and a core contributor to Project NANDA, the MIT-born initiative building an open internet for autonomous agents. She develops infrastructure that helps independently operated agents discover one another, verify identities, exchange data, and coordinate without relying on a single platform.
Gorskikh graduated from Boston University’s Questrom School of Business in 2023. There, she co-founded DREAM Venture Labs, a nonprofit pairing immigrant and refugee entrepreneurs with student volunteers who assist with business formation, branding, accounting, and strategy. The venture won Boston University’s 2023 Refugee Challenge.
Her subsequent work expanded into decentralized agent infrastructure. She co-authored 2025 research on agent discovery and identity, served as chief technology officer of the AI-visibility company Citable, and helped teach MIT’s 2026 course on networked AI agents. Through Maritime, she developed sleep-and-wake agent hosting, which reduces the cost of running many agents by suspending idle compute.
- Agent discovery needs verifiable identity. The NANDA Index and AgentFacts combine agent directories with signed records describing capabilities, permissions, communication endpoints, and provenance. Requests can resolve differently depending on the requester’s identity and authorized access.
- Agents need trustworthy data, too. Her co-authored Data Facts research proposes linking agent identities to structured dataset descriptions, including access restrictions, freshness, endpoints, and integrity information.
- Robots belong on the agentic web. In her essay on robotics and agent frameworks, Gorskikh describes Raspberry Pi robot-swarm experiments and treats cameras, grippers, microphones, and wheels as callable tools. Time-sensitive control remains local; planning and coordination can run through cloud-based agents, with safety checks before instructions reach hardware.
- Coordination must be tested under realistic conditions. Her AI Engineer conference appearance details NandaTown, a simulation environment for agent discovery, messaging, negotiation, auctions, voting, consensus, and recovery from failures. Scripted agents establish controlled baselines before experiments incorporate actual AI models.
