Member of Technical Staff, Computer-Use Agents

About Hark

Hark is an artificial intelligence company building advanced, personalized intelligence. One that is proactive, multimodal, and capable of interacting with the world through speech, text, vision, and persistent memory.

We're pairing that intelligence with next-generation hardware to create a universal interface between humans and machines. While today's AI largely operates through chat boxes and decade-old devices, Hark is focused on what comes next: agentic systems that interact naturally with people and the real world.

To get there, we're developing multimodal models and next-generation AI hardware together - designed from the ground up as a single, unified interface for a new era of intelligent systems.

About the Role

The Computer-Use Agent (CUA) team at Hark builds agents that use a computer effectively and efficiently to accomplish increasingly long-horizon tasks. These are the models that let Hark get real work done on a user's behalf, from a single errand to workflows that run for hours.

We're looking for engineers and researchers who are motivated to build the best computer-use agents in the world. This role demands high-quality engineering combined with genuine research ability, and the agency to investigate promising novel algorithms and approaches that will shape the next generation of computer-use agents at Hark.

You'll work collaboratively in a fast-moving, high-performance team across every aspect of training computer-use agents. We expect deep technical expertise in one or more of the following areas: data and environment engineering, agentic evals, RL post-training, and training and inference infrastructure for agents.

This is a high-ownership role on a small team. Beyond the CUA team, you'll work closely with the harness and product, pretraining, and infrastructure teams.

Responsibilities

The ideal candidate brings exceptional depth in one or more of the areas below, paired with a working understanding of the rest.

• Environments & Data: Build and scale interactive environments and task generators for computer use, including realistic applications, stateful workflows, verifiable success checks, and synthetic trajectories for SFT and RL. • Evaluation: Design benchmarks and eval harnesses that measure task success, reliability, efficiency, and safety on long-horizon tasks, and turn model failures into concrete data, environment, or training fixes. • Post-Training: Develop and scale SFT, RL (PPO/GRPO and beyond), reward modeling, and grader and verifier techniques that make agents both more capable and more efficient, including in long-horizon and sparse-reward settings. • Research & Novel Methods: Identify promising new algorithms and approaches, prototype them quickly, and test them rigorously with strong baselines and ablations to decide what goes into the next major model run. • Data Quality: Build filtering, curation, difficulty calibration, and curriculum methods so every training run gets the highest-signal data we can produce. • Agent Infrastructure: Build and scale the training and inference infrastructure behind large-scale agent rollouts and RL, improving throughput, latency, reproducibility, observability, and cost. • Production Impact: Partner with harness, product, and infrastructure teams to close the loop between real user tasks and model improvements, and debug hard failures in shipped models.