Examining how a principal can effectively \emph{delegate} tasks to an agent is a central problem in many real-world settings, studied extensively in economics, computer science, and operations research.
Representative examples include a team manager delegating problem-solving to members, and a funding agency selecting researchers to advance specific fields.
It subsumes a broader class of applications in classic economy where a public sector delegating trade governance to a self-interested broker, and modern digital economy such that a platform like YouTube relying on creators for content production.

In each case, the principal benefits from the agent’s information advantage, lacking the expertise or time to decide alone. However, agents often have ulterior motives misaligned with the principal’s, creating a \emph{moral hazard} where the agent may act strategically for personal gain. The principal’s challenge is to design incentives that align the agent’s behavior with the principal’s objectives.

This article explores two fundamental questions in this context: (i) given a self interested agent, how could the principal design an efficient mechanism that effectively incentivizes the agent's behavior in favor of the principal, and what would be the corresponding approximate guarantees to the first-best (omniscient) outcome? and (ii) if the principal cannot intervene in the agent’s decision, how much worse the outcome can be compared to the first-best outcome in hindsight?

My research investigates these questions in four representative domains spanning classic and modern applications: (1) delegation in organizations, (ii) delegation in auctions, (iii) delegation in modern online platforms, (iv) and delegation in the era of generative AI, offering both theoretical insights and practical relevance in the broader literature of algorithmic game theory.The results are based on part of my papers published at STOC, EC, SODA, NeurIPS, ICML, AISTATS, AAAI as well as working papers.