Agent Runtime Engineer

Objective

Build the execution path that gives every supported agent action an enforceable decision and an inspectable result.

What you’ll own

• The action-time gate between agents and the systems they call
• Policy evaluation for identity, scope, approvals, and tool access
• Sandboxing, egress allowlists, and credential handling for tool calls
• Stop, pause, and quarantine paths for running agents

Evidence of success

• Every supported tool call has an attributable allow, wait, or deny decision
• A paused or quarantined agent cannot continue through another execution path
• Policy and runtime regressions fail a release before they reach customers

Constraints you should know

• The gate sits on a latency-critical request path
• Visible controls must enforce the behavior they describe
• Retries and alternate tools must not bypass a denied decision

Working style

• Work in person with the founding team in San Francisco
• Own production behavior and join the on-call path for what you ship
• Use Rust on the current hot path; bring Go or equivalent systems depth

What we look for

• Shipped a latency-critical control plane or policy engine in production
• Production Rust, Go, or equivalent on a hot request path
• Built sandboxing or capability isolation exercised by real failures or adversaries
• Owned on-call for a system whose outages wake people up