Software Engineer - Agent Infra

Details

Perplexity is an AI-powered answer engine founded in December 2022. The company builds accurate, trustworthy AI to power decision-making and assistive AI. Perplexity answers hundreds of millions of queries per month and is rapidly growing.

This role is for a Software Engineer on the Agent Infrastructure team. You will build core infrastructure that powers agentic products across Perplexity (Search, Deep Research, Comet browser). The work includes designing and evolving runtime, orchestration, evaluation, and training systems that enable agents to plan, use tools, browse, and verify at scale.

Responsibilities

• Design and implement a highly reliable and scalable agent runtime: orchestration, shared state and memory, tool calling interfaces, and scheduling for cost, latency, and quality.
• Build secure sandboxed execution for agent actions and code; optimize cold start, isolation, and observability.
• Ship unified interfaces for multiple model sizes and providers. Integrate with open tool ecosystems such as MCP-style connectors for data and actions.
• Develop an evaluation platform for online and offline assessments, A/B tests, safety checks, and regression gates to improve agent reliability over time.
• Partner with Research, Inference, and Search to land new agent capabilities end to end, from prototype to production.

Requirements

• 6+ years of industry experience building large-scale systems or platforms.
• Experience building agent applications with tool calling, context engineering, or open connector integrations.
• Strong coding skills in one or more of: Python, Java, Go. Comfortable with service design, APIs, and data models for high-throughput systems.
• Working knowledge of containers, virtualization, and sandboxing. Familiar with metrics, tracing, on-call, and incident practices.
• Bias to own problems across layers, collaborate in fast-moving teams, and ship.

Nice to have

• Prior work on agent orchestration pipelines: task routing, planning, memory graphs, vector search, or browser automation.
• Experience with evaluations, preference optimization, or RL to improve agent reliability.
• Exposure to MCP or similar open protocols for tool and data connectivity.
• Security hardening for multi-tenant workloads, snapshotting, or untrusted code execution.

Compensation & Benefits

• Cash compensation range: $200,000 - $300,000 per year. Final offers vary based on experience and expertise.
• Equity may be part of the total compensation package.
• Benefits include comprehensive health, dental, and vision insurance, and a 401(k) plan.