Saaket Agashe

Curriculum Vitae
Google Scholar
Email Address
GitHub Profile
LinkedIn
Twitter

I am a third-year Ph.D. student at UC Santa Barbara, advised by Professor Xin (Eric) Wang. I previously earned an M.S. in Computer Science from UC Santa Cruz and a B.Tech. in Electronics Engineering from VJTI Mumbai.

My research focuses on building and evaluating LLM agents that can plan and act over long horizons in real-world environments. I am particularly interested in developing sample-efficient methods that enable agents to learn from minimal but direct interactions with their environment.

For more details, check out my CV or drop me an email.

Announcements

Research

[1] Context Bootstrapped Reinforcement Learning
Saaket Agashe, Jayanth Srinivasa, Gaowen Liu, Xin Eric Wang
Preprint, 2026
Description: CBRL addresses exploration inefficiency in RLVR by injecting few-shot demonstrations into training prompts with a curriculum that anneals to zero, forcing the model to internalize reasoning patterns into its weights. We show consistent gains across two model families and five reasoning tasks, including domains where the model initially has near-zero success rate.
Paper

[2] Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents
Saaket Agashe, Kyle Wong, Vincent Tu, Jiachen Yang, Ang Li, Xin Eric Wang
COLM 2025
Description: Agent S2 is a compositional framework for computer use agents that delegates cognitive responsibilities across various generalist (planning) and specialist (grounding) models.
Website Paper Code

[3] Self-Resource Allocation in Multi-Agent LLM Systems
Alfonso Amayuelas, Jingbo Yang, Saaket Agashe, Ashwin Nagarajan, Antonis Antoniades, Xin Eric Wang, William Wang
Preprint
Description: We study how LLM-based agents allocate tasks in multi-agent systems, comparing planner and orchestrator strategies for efficiency, validity, and agent utilization.
Paper

[4] Agent S: An Open Agentic Framework that Uses Computers Like a Human
Saaket Agashe, Jiuzhou Han, Shuyu Gan, Jiachen Yang, Ang Li, Xin Eric Wang
ICLR 2025
Description: Agent S integrates experience-based learning, web retrieval, and hierarchical planning for OS-level GUI automation, achieving state-of-the-art performance on the OSWorld benchmark.
Website Paper Code

[5] LLM-Coordination: Evaluating and Analyzing Multi-agent Coordination Abilities in Large Language Models
Saaket Agashe, Yue Fan, Anthony Reyna, Xin Eric Wang
Findings of NAACL 2025
Description: Introduced the LLM-Coordination Benchmark and conducted a comprehensive analysis of LLMs in Pure Coordination Games.
Website Paper Code

[6] Localization using Spatial Descriptions
Saaket Agashe, Xin Eric Wang
Description: Visually grounding Spatial Descriptions to a single point in an image based on descriptions of objects in the neighborhood of the point.
Code

[7] Interacting with Next-Phrase Suggestions: How Suggestion Systems Aid and Influence the Cognitive Processes of Writing
Advait Bhat, Saaket Agashe, Niharika Mohile, Parth Oberoi, Ravi Jangir, Anirudha Joshi
IUI 2023
Best Paper Honourable Mention
Description: Exploratory qualitative study to understand how writers interact with next-phrase suggestions.
Code Paper

[8] How do people interact with biased text prediction models while writing?
Advait Bhat, Saaket Agashe, Anirudha Joshi
HCI+NLP Workshop, EACL 2021
Description: Pilot study to understand how people interact with next phrase suggestion system.
Code Paper

[9] Perception and Motion Planning for Autonomous mobile manipulator
Saaket Agashe, Akshay Paralikar, Shweta Kumaran, Shambhavi Kuthe, Aditya Gawali, Hiten Kothari, Shashank Deshmukh, Faruk Kazi
Description: Develop a functional autonomous mobile manipulator for testing algorithms developed in simulation in the real world.
Code