Data Operations Manager - Computer Use & Tool Use

Data Operations Manager – Computer Use & Tool Use

Join us to apply for the Data Operations Manager – Computer Use & Tool Use role at Anthropic
.

About Anthropic – Our mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.

As Data Operations Manager for Computer Use & Tool Use, you’ll build and scale data operations that advance Claude’s computer use capabilities and tool use safety. You’ll partner with research teams to design and execute data strategies, manage vendor relationships, and own the entire data pipeline from requirements to production. This is a zero‑to‑one role requiring technical depth to understand what makes high‑quality training data for autonomous agents, but your focus will be on strategy and execution rather than hands‑on engineering.

Think of a technical founder who evolved from writing code to building the business.

The data strategies and operations you build will directly determine how well Claude can use tools safely, operate computers autonomously, and maintain quality across long‑horizon agentic workflows. You’ll work with world‑class researchers advancing frontier capabilities, safety, and model performance while building the operational infrastructure to scale these efforts. We’re looking for someone who gets excited about scaling quality for complex, multi‑turn agent interactions – someone who can think strategically about data needs for both capabilities and safety, build the right partnerships, and execute flawlessly.

• Develop and execute data strategies for computer use, tool use safety, and agentic AI research.
• Partner with research leaders to translate technical requirements into operational frameworks.
• Build data collection and evaluation systems for complex scenarios: prompt injection robustness, multi‑turn agent conversations, adversarial attacks, autonomous workflows.
• Scale the generation of realistic evaluation environments that capture real‑world tool use and computer use challenges.
• Identify, evaluate, and manage specialized contractors and vendors for technical data collection.
• Implement quality control processes to ensure data meets training requirements for both capabilities and safety.
• Manage multiple complex projects simultaneously, balancing research velocity with rigorous evaluation standards.
• Track metrics and communicate progress to stakeholders.

• Have 3+ years in technical operations, product management, or entrepreneurial experience building from zero to scale.
• Have strong technical foundations – proficiency in Python and understanding of ML workflows, RL environments, and evaluation frameworks.
• Have strong communication skills and can effectively engage with both technical and non‑technical stakeholders.
• Are familiar with how LLMs work and could describe concepts such as RLHF, tool use, and agentic workflows.
• Understand the unique challenges of evaluating autonomous systems and long‑horizon agent behaviors.
• Are highly organized and can manage multiple parallel work streams effectively.
• Have a high threshold for navigating ambiguity and can balance strategic priorities with rapid execution.
• Thrive in fast‑paced research environments with shifting priorities and novel technical challenges.
• Are passionate about AI safety and understand the critical importance of high‑quality data in building safe, capable agentic systems.

• Experience at companies training AI models, building AI agents, or creating AI training data, evaluations, or environments.
• Knowledge of computer and tool use safety challenges like prompt injection, data exfiltration attempts, or adversarial attacks.
• Experience with RLHF, reinforcement learning techniques, or similar human‑in‑the‑loop training methods.
• Domain expertise in computer use automation, security, or AI safety evaluation.
• Familiarity with model performance monitoring, training…