Visiting Engineer: Enterprise Agent Infrastructure

career•Mountain View, CA

10d•Onsite

About The Position

Every large company runs on internal systems that are mission-critical but brittle — custom admin panels, ETL pipelines, dashboards, and operational scripts held together by institutional knowledge and hope. This role is to build AI agents that operate behind the enterprise firewall, automating the operational tasks that currently require manual intervention or one-off scripts by securely connecting to internal databases, APIs, and legacy systems.

Requirements

Strong backend engineering skills. You have built and shipped production systems that handle real data at real scale.
Deep experience with enterprise infrastructure: databases, APIs, authentication systems, networking constraints.
Hands-on experience building with LLMs, particularly agentic architectures, tool use, and function calling.
Understanding of security and compliance requirements in enterprise environments.
Ability to work autonomously and make product and architecture decisions without waiting for direction.

Nice To Haves

Experience with MCP (Model Context Protocol) or similar tool-use frameworks for AI agents.
Experience building internal developer platforms or internal tooling at scale.
Background in systems integration, middleware, or ETL pipeline development.
Founder or founding engineer experience building infrastructure products.

Responsibilities

Design the core agent architecture for operating in constrained network environments where cloud-first assumptions break down.
Build connectors for common enterprise data sources: SQL databases, internal APIs, file shares, legacy mainframe interfaces.
Develop a natural language task definition layer that translates operator intent into safe, auditable multi-step agent actions.
Implement security primitives: credential vaulting, least-privilege execution, action sandboxing, and rollback capabilities.
Work with enterprise design partners to validate the agent framework against real internal tooling pain points.