Founding Engineer — AI Agent Security

Greylock Partners; San Francisco (CA); On-site; USD 180,000 - 280,000; Full time

Job summary

Greylock Partners is seeking a hands-on founding engineer in San Francisco to build production software for AI agent security. You will own ambiguous engineering problems, from design through deployment, and work closely with the founders on technical direction and execution.

This role focuses on runtime systems, model/inference integrations, and production workflows that securely read untrusted information, use tools, and act across real environments.

Qualifications

• Personal experience building and operating substantial production systems.
• Experience building software for models, agents, or inference in production.
• Ability to take ambiguous engineering problems from scope to deployment and iteration.

Responsibilities

• Build runtime systems that evaluate an agent’s context before tool use.
• Develop model/inference integrations with latency and reliability requirements.
• Own production architecture, observability, isolation, deployment safety, and recovery.
• Collaborate with founders on technical direction and execution.
• Design and ship production workflows that connect user intent, content, tool calls, and agent actions.

Skills

Production systems

Agent software

Inference optimization

Model training

Ambiguity handling

Job description

Seed-stage company with fewer than 10 people in San Francisco building an adaptive security layer for autonomous AI agents.

As agents gain access to company data, credentials, and tools, permissions alone cannot determine whether an action serves the user’s goal. This team is building a runtime firewall that evaluates an agent’s intent, context, and actions, blocks harmful behavior, and learns from newly discovered attacks.

The team works closely together in person in San Francisco. Its systems already protect production agent workflows.

Why this role

This is a hands-on founding engineering role building production software for agent security.

The hard problem is making fast, reliable security decisions as agents use tools and data, then turning attacks and production evidence into better defenses. You’ll own ambiguous engineering problems across the product and carry solutions from design through deployment and operation.

You’ll work directly with the founders on technical direction and execution.

What you’ll build

Build and own foundational software that protects AI agents as they read untrusted information, use tools, and act across real environments.

Your strongest area may be inference, models, or agent harnesses. You’ll apply that depth while building the runtime systems, integrations, evaluations, and production workflows that make the product work reliably.

What you’ll work on

• Runtime systems that evaluate an agent’s context and actions before it uses tools
• Model and inference integrations with strict latency and reliability requirements
• Agent and attack-generation harnesses that expose failures and make them reproducible
• Evaluation and learning loops built from verified attacks and production outcomes
• Software that connects user intent, encountered content, tool calls, and an agent’s sequence of actions
• Production architecture, observability, isolation, deployment safety, and recovery
• AI-native engineering workflows that increase output while preserving rigorous review and testing

What they’re looking for

• Strong software engineering fundamentals demonstrated by personally building and operating substantial production systems
• Experience building software for models, agents, or inference in a production setting
• Ability to take an ambiguous engineering problem from initial scope through implementation, deployment, and iteration
• Judgment about correctness, latency, and reliability when the system’s decisions have real consequences
• Range to work across product behavior, systems, and AI components as the problem requires
• Initiative and clear communication in a small team where priorities evolve quickly

Strong signals

• Owned an AI or agent system from design through production use and iteration
• Brought unusual depth in inference optimization, agent harness engineering, or model training to a shipped system
• Solved meaningful engineering problems beyond model training or evaluation while shipping an AI product
• Turned an unclear technical problem into a working product or durable system
• Uses coding agents, automated evaluation, testing, and review to expand engineering output without losing command of the underlying work
• Security experience is valuable, but direct experience in agent security is not required

Level and experience

This is a hands-on individual contributor role for engineers with meaningful production software ownership. The team will consider experienced engineers and exceptional early-career builders. The scope and results of your work matter more than a particular title or number of years.

The role is based in San Francisco and is intended for someone who wants to build closely with the team in person.

About Us

Greylock is an early-stage investor in hundreds of remarkable companies including Airbnb, LinkedIn, Dropbox, Workday, Cloudera, Facebook, Instagram, Roblox, Coinbase, Palo Alto Networks, among others. More can be found about us here: https://greylock.com/

Please note: We are not recruiting for any roles within Greylock at this time. This job posting is for direct employment with a startup in our portfolio.