NanoCo ships managed enterprise service: one sandboxed AI agent per employee

NanoCo, the Tel Aviv startup behind the open-source NanoClaw agent framework, announced a managed enterprise service on May 20 that provisions one sandboxed agent per employee. The New Stack's coverage frames the architecture as a deliberate bet: rather than shared agent infrastructure governed by team or org policy, NanoClaw's commercial pitch is per-employee isolation — each agent runs in its own sandbox, with its own credentials, its own audit trail, and its own resource budget.

The technical reasoning is governance-driven. Multi-tenant agent platforms struggle when one employee's agent leaks context into another's, or when an agent acts with org-wide privileges and trips a least-privilege policy. Per-employee sandboxing mirrors how enterprise IT thinks about user-account isolation: each human gets a machine identity, each agent inherits a scoped subset of that identity, and audit is per-user. NanoClaw exposes the framework as open source and monetizes the managed orchestration layer.

This ships into a crowded but increasingly serious agent-governance week. AWS published its Cedar-policy rationale for Bedrock AgentCore. Anthropic's Claude managed-agent MCP tunnels and self-hosted sandboxes hit research preview on May 19. NanoCo's pitch is a different cut — instead of 'sandbox the runtime' (Anthropic) or 'policy-gate the actions' (AWS), it's 'partition the identity.' All three converge on the same problem: agents that act on enterprise data and APIs need governance primitives that don't yet exist in standard IAM.

For a Tel Aviv startup with an open-source framework, the commercial bet is reach via NanoClaw and revenue via managed deployment. The skeptical read is that per-employee sandboxing implies linear cost scaling with headcount — which is fine for a 100-employee design partner but expensive at 100,000. Watch for which design partners NanoCo discloses and how its per-seat pricing lands versus Claude Managed Agents and AgentCore.