AI Agency: Paths, Uncertainties and Solution Spaces

One great focus in the Generative AI industry is advancing the abilities of AI systems to have impact in the world. Currently, Generative AI systems primarily produce online digital outputs based on their own training data and generative capabilities. As we have seen, these capabilities are quite powerful, enabling users to enjoy enhanced creativity and productivity at high levels of proficiency. Given the power and reach of LLM-based foundation models to find sensible answers to sophisticated questions posed, one of the next horizons is expanding the ability of AI systems to have autonomous decision making abilities with farther reaching impact in the world. That is, one key goal of the leading foundation models players as well as numerous start-ups is to build AI agents that can execute complex tasks in both the digital and physical worlds. The digital and physical worlds each present different product design and technical execution challenges. Each also presents overlapping but distinct legal challenges relating to risk and potential liability. The prospect of building effective AI agents to operate at scale for thousands if not millions of users will require a myriad of problem solving. The goal of this article is to frame the goals of agentic systems, flush out the various ways that AI agents can be implemented to have far reaching impact in the world, and outline some of the current challenges that must be addressed to make such prospects a reality.