The company builds infrastructure that lets other companies run their own work with AI agents, and it starts with trust. Its first product is an MCP gateway that gives agents controlled access to company data and tools.
Before agents can reach systems like Slack, GitHub, Jira, Snowflake, internal databases, or production, a company needs to know what each agent can access, what it did, and why it was permitted. The gateway narrows an agent's access to what its task requires and records the data flows and audit logs that security teams rely on. The company plans to add connectors, monitoring, sandboxing, long-running agents, and memory, with the aim of becoming the system a company's work runs on.
The company is AI-native. Its staff use Claude and Codex daily, and it runs more agents than people. These agents work in Slack with the team across sales, marketing, deal operations, and engineering, using the same gateway the company sells to customers. It is building a full company operating system around them.
The company launched in February 2026 and now has more than 50 paying customers, with strong inbound demand. The founders are machine learning researchers from Stanford, Berkeley, and CMU, with applied research at Google Brain and Waymo, and one of them co-founded Coursera.
Responsibilities
Take features from idea to production yourself, across the gateway, enterprise deployments, and agent systems.
Work mainly on backend and infrastructure, and also build UI or fix connectors when the product or a security review needs it.
Use Claude and Codex as your main tools, along with a group of agents that write much of the code and review pull requests.
Make the decisions agents cannot: how to frame problems, architecture, product and UX choices, hardening enterprise paths, designing connectors, protecting security-sensitive areas, and spotting generated code that is subtly wrong.
Ship within hours and learn directly from the founders and customers.
Requirements
You have shipped to real users, through a product, an internal tool, an open-source project, or a serious side project. You ship and then iterate without waiting for specs or sign-off.
You have solid fundamentals and learn quickly. Speed of learning counts as much as years of experience.
You put product value and customer needs ahead of technical purity.
You take infrastructure seriously and consider blast radius, trust boundaries, and the effect of your code on every customer's agents.
You can read code you did not write, reason from first principles, and explain the tradeoffs you choose.
You work comfortably across the stack in TypeScript, Python, and CockroachDB, or you learn new languages, protocols, and domains quickly.
You are enthusiastic about building with agents and can tell when they help and when they hide a problem.
You already build with AI every day and want room to work without heavy process.
Experience with developer-facing products, protocol or gateway work, or infrastructure at scale.
Background in security, identity, or permission systems.
Contributions to open source, MCP servers, or agent frameworks.
Design sense, whether you are an engineer who cares about design or a designer who thinks about infrastructure.
Conditions
The role is in person in the San Francisco Bay Area.
The team is flat and the market moves quickly, so you decide a lot about what you work on and how.
The company hires at every level, from early-career builders to experienced engineers, and sets level and pay to match each person.
Benefits
Competitive pay and meaningful equity.
Wide scope and ownership over features that ship to enterprise customers.
Building and running agents on the same platform the company sells.
Direct access to the founders, customer context, and product direction, plus mentorship as you grow.
Health benefits, on a team that puts family and health first.