Yucheng Jiang

Ph.D. student in Computer Science, Stanford University

I am a Ph.D. student in Computer Science at Stanford University, advised by Monica S. Lam. My research lies at the intersection of natural language processing and human–computer interaction. I study AI agents that learn from human expertise. In turn, these agents help people do more valuable scientific and economic work.

Previously, I received my M.S. in Computer Science from Stanford, worked as a software engineer at Apple, and received my B.S. in Statistics and Computer Science from the University of Illinois Urbana-Champaign, where I was supervised by Jiawei Han.

Portrait of Yucheng Jiang
01

Deep research agents

Deep research agents answer complex, open-ended questions by searching the web and large databases, cross-checking what they find, and writing long-form reports with citations. They turn hours of searching and reading into a report that people can verify, learn from, and build on.

  1. COLM 2026

    DataSTORM

    Deep research over large-scale databases

    An agent that applies exploratory data analysis and data storytelling to large databases. It forms hypotheses, verifies them with SQL, web, and private-corpus search, and writes a fact-checked analytical report.

  2. EMNLP 2024

    Co-STORM

    Learning the unknown unknowns through collaborative agent discourse

    Helps people learn about unfamiliar topics through a simulated roundtable conversation among LLM agents, which they can follow or join at any point. The agents raise questions people would not think to ask, making serendipitous discovery more likely, and a dynamic mind map organizes the discussion to reduce mental effort over long sessions.

  3. NAACL 2024

    STORM

    Writing grounded, Wikipedia-style articles from scratch

    Writes cited long-form articles by modeling the pre-writing stage: discovering perspectives on a topic, simulating expert interviews grounded in web search, and curating the findings into an outline.

02

Human–agent collaboration and agents for real work

AI agents are beginning to take on real work, but most are designed to act alone. This work studies how people actually carry out their tasks and what they want from AI, and builds agents that collaborate with people and keep them in control.

  1. EMNLP 2026

    Task Model Induction

    Learning how work is done from computer-use traces

    Induces symbolic, auditable models of how people perform tasks from passively recorded computer use (screenshots, mouse, and keyboard events), capturing both the goals they pursue and the procedures they follow.

  2. ICLR 2026

    Collaborative Gym

    Enabling and evaluating human–agent collaboration

    A framework for agents that work alongside people in shared task environments through asynchronous, non-turn-based interaction, evaluated with both simulated and real users.

  3. Preprint 2025

    Future of Work with AI Agents

    Automation vs. augmentation across the U.S. workforce

    A worker-centric audit of which tasks people want AI agents to automate or augment, compared with what current agents can do. Introduces the Human Agency Scale and the WORKBank dataset.

    • Featured in Stanford HAI's 2026 AI Index, Microsoft's New Future of Work Report, and Allianz Research
    • Reported by Forbes, Fast Company, ZDNET, and Tencent News
    PaperWebsiteDataset
03

Agent context management

Context management is the part of an agent harness that decides what information the model sees at each step. It lets agents persist through long-horizon tasks, so that the reasoning ability of today's models carries into sustained, long-running work.

  1. ACL 2026 Findings

    STITCH & CAME-Bench

    Grounding agent memory in contextual intent

    STITCH indexes each step of a long agent trajectory by its contextual intent (goal, action type, and key entities) and retrieves memory by intent. CAME-Bench evaluates memory in long interactions where goals shift and entities recur.

Talks

  1. Stanford CS 224V

    Deep Research Agents

  2. Stanford HAI · Public AI Assistant to World Wide Knowledge

    Human–AI Co-Research: Interaction and Collaboration in Public AI Assistants

  3. EMNLP 2025

    Creative Planning with Language Models

  4. DeepLearning.AI

    Deep Research and Automatic Educational Content Creation

  5. Stanford CS 224V

    Knowledge Curation and Human–AI Interaction

  6. Google DeepMind

    Co-STORM: Engaged Human Learning through Collaborative Discourse with LLM Agents

  7. EMNLP 2024 Workshop on NLP for Wikipedia

    Co-STORM: Engaged Human Learning through Collaborative Discourse with LLM Agents

  8. Synopsys

    Deep Research for Industry: Educating Users and Driving Vertical Adoption

  9. Stanford CS 224V

    Knowledge Curation

Teaching

  1. Stanford University

    CS 224V: Conversational Virtual Assistants with Deep Learning

  2. University of Illinois Urbana-Champaign

    CS 222: Software Design Lab

  3. University of Illinois Urbana-Champaign

    CS 225: Data Structures

  4. University of Illinois Urbana-Champaign

    CS 374: Introduction to Algorithms and Models of Computation

  5. University of Illinois Urbana-Champaign

    CS 126: Software Design Studio

  6. University of Illinois Urbana-Champaign

    CS 128: Introduction to Computer Science II

  7. University of Illinois Urbana-Champaign

    CS 125: Introduction to Computer Science