Discovery
How can AI agents search, hypothesize, and optimize across science, engineering, and infrastructure?
Overview
AI agents are increasingly used to search over code, experiments, and designs to produce candidate discoveries. This workshop focuses on agents that operate beyond benchmarks — under expensive evaluations, noisy measurements, and real deployment constraints.
We invite work on discovery agents in science, engineering, and infrastructure, where validation is hard and human oversight matters.
Evaluation
How do you validate agent-driven discoveries when ground truth is limited and experiments are expensive?
Deployment
What breaks when discovery agents move from benchmarks to real-world settings?
Topics of Interest
We welcome submissions on building, evaluating, and deploying AI agents that operate under real-world constraints. These areas are representative, not exhaustive.
-
Search and Optimization with LLM Agents
Agents that explore design spaces, generate candidate solutions, and optimize over code, configurations, or experimental parameters — including evolutionary, iterative, and tool-augmented approaches.
-
Agents for Systems Infrastructure
AI agents applied to scheduling, capacity planning, database tuning, network architecture, cluster management, performance debugging, and other infrastructure operations.
-
Agents for Scientific Discovery
Agents that assist with hypothesis generation, experiment design, proof construction, data analysis, or literature synthesis — across domains such as mathematics, natural sciences, and engineering.
-
Evaluation Under Real-World Constraints
Methods for validating agent behavior when ground truth is limited, experiments are expensive, feedback is noisy, or reliability requirements are strict.
-
Deployment Reports and Failure Analysis
Case studies from applied settings, including adoption challenges, failure modes, operational lessons, and what didn't work.
-
Human-Agent Collaboration and Oversight
Workflows where domain experts direct, audit, or intervene in agent-driven processes — including trust calibration, delegation boundaries, and expert-in-the-loop design.
Call for Papers
We invite submissions on AI agents for discovery in real-world settings. Deployment reports, case studies, and lessons from applied systems — including failures — are especially welcome.
Paper submission
Submission
4-page short papers and 9-page long papers, using the official paper format.
OpenReview link coming soon
Paper Template
Important dates · All deadlines follow the Anywhere on Earth (AoE) timezone.
Submission deadline
May 1, 2026
Notification
May 13, 2026
Camera-ready
May 24, 2026
Workshop day
May 26, 2026
Submission details
- Formats: 4-page short papers and 9-page long papers.
- File format: Single PDF, up to 50 MB.
- Template: Official paper template.
- Length: References and appendices do not count toward the page limit.
- Review: Double blind, with at least two reviews per submission.
- Criteria: Relevance, technical quality, novelty, and discussion potential.
Presentation and policies
- Presentation: All accepted papers will be presented as posters, and a small number will be selected for oral presentations.
- Award: The workshop will present a Best Paper Award.
- Anonymization: Review PDFs must not reveal author identity, including linked or supplementary material.
- Non-archival: Concurrent submissions and recently accepted work are welcome if venue policies permit.
- Visibility: Submissions and reviews will not be public; only accepted papers will be made public.
Program Committee
Lakshya AgrawalUC Berkeley
Batu ElStanford
Mert CemriUC Berkeley
Alex DimakisUC Berkeley / Bespoke Labs
Shu LiuUC Berkeley
Tentative Schedule
May 26, 2026 · San Jose, CA · subject to change
8:30
Opening Remarks
8:45
Invited Talk 1
9:15
Invited Talk 2
9:45
Paper Session
10:15
Coffee Break
10:30
Invited Talk 3
11:00
Poster Session 1
11:45
Lunch
13:30
Invited Talk 4
14:00
Invited Talk 5
14:30
Invited Talk 6
15:00
Panel Discussion
15:45
Coffee Break
16:00
Poster Session 2
16:45
Open Discussion
17:15
Closing Remarks
Organizers
Shubham Agarwal
UC Berkeley
Lakshya Agrawal
UC Berkeley
Mert Cemri
UC Berkeley
Alex Dimakis
UC Berkeley / Bespoke Labs
Batu El
Stanford
Alex Krentsel
UC Berkeley
Eric Liang
Databricks
Shu Liu
UC Berkeley
Rui Meng
Sylvia Ratnasamy
UC Berkeley
Ion Stoica
UC Berkeley
Matei Zaharia
UC Berkeley / Databricks
Sponsors
Bespoke Labs
Bespoke Labs is a sponsor of this workshop.
Contact
For sponsorship inquiries or any workshop-related questions, contact cais26.ai.discovery@gmail.com.