Felicis AI Investments

A browser task can be bounded by pages, sessions and web interactions, while computer use may cross application and operating-system boundaries. Permissions and state changes therefore need to be assessed in context. A successful demonstration in one environment does not establish that the same agent can safely handle every desktop or browser workflow.

Traditional automation often follows predefined steps or selectors. An agent can use a model to interpret a goal and choose actions, which adds flexibility but also uncertainty. Evaluation should examine successful task completion, permissions, state changes and recovery from errors, rather than relying only on demonstrations of a short browsing sequence.

Actions in business software can change files, send messages or affect external systems. Approval boundaries help assign responsibility for consequential changes, especially when a model may misunderstand context. The appropriate boundary depends on the task and permissions; an agent’s ability to perform an action does not itself establish authorization to perform it.

Specialist judgment can help define tasks, evaluate outputs and identify failures that generic benchmarks miss. That work differs from simply supplying large volumes of undifferentiated data. Mercor’s inclusion makes the human contribution relevant to this portfolio view, without implying that every assignment serves the same training or evaluation purpose.

Evaluation should consider control, consistency, editing workflow, rights and the cost of producing usable outputs. A compelling short clip does not establish that a tool fits a production pipeline. Runway’s portfolio inclusion identifies an investment relationship; specific performance claims still need independent product-level evidence.

Expert work can contribute judgment, task definition and evaluation, while a creative tool helps users produce and revise media. Both can benefit from AI adoption without serving the same buyer or replacing the same activity. The distinction matters when assessing demand, the role of people and what successful output actually means.