About the company
Perplexity Computer is one of the defining products of the new era of agentic AI. Millions of people use Perplexity to transform knowledge into action, and the Agent Capabilities team sits at the intersection of frontier AI research and product innovation.
Responsibilities
- Evaluate frontier models against real user tasks, identify useful behaviors and failure modes, and turn the most promising advances into production agent systems
- Own the lifecycle from rapid prototyping and evaluation through launch, monitoring, and iteration
- Improve agents' ability to plan, use tools, manage context, recover from errors, and complete long-running tasks reliably
- Apply state of the art ML and LLM techniques to design scalable agent capabilities such as skills, plugins, artifact generation, tools integration, auto-research, and multi-agent collaboration
- Own agent behavior and capabilities end-to-end, from user-facing products and interfaces to backend services
- Define offline and online evaluations for task completion, correctness, safety, latency, cost, and user satisfaction
- Build secure, observable, and reliable agent systems, including permissions and safeguards for sensitive actions
- Develop tracing, replay, and monitoring infrastructure
- Collaborate closely with PM, Data Science, Research
Requirements
- Typically 6+ years of professional software engineering experience
- Strong software engineering fundamentals, with experience building and operating AI/ML products, backend services, or distributed systems at scale
- Experience owning the AI product lifecycle, including data analysis, rigorous evaluation, production monitoring, and iterative improvement
- Practical experience in one or more relevant areas, such as agent harnesses, tool use, context engineering, model evaluation, browser automation, or long-running task execution
- Strong product judgment and execution
- Genuine interest in frontier AI capabilities, agent systems
Nice to have
- Experience with LLM context engineering or harness engineering, subagents, coding assistants, long-running or autonomous task execution
- Deep familiarity with the strengths and limitations of current model families
- Experience building agent permissions, safeguards, evaluation infrastructure, or production observability systems
- Experience with mid-training, post-training, or reinforcement learning for frontier or open-source models
- AI/ML research experience demonstrated through publications, open-source contributions
- Time spent at a fast-growing startup or on a high-ownership engineering team