Registration is now open! Register now

Wednesday Schedule

Wednesday, May 27 — Main Conference Day 1

All sessions take place at the DoubleTree by Hilton San Jose, 2050 Gateway Place, San Jose. Click any paper session to see the individual talks.

7:30 AM 8:30 AM

AM Break

Bayshore Foyer

9:00 AM 10:15 AM

Welcome & Keynote 1 — Andy Konwinski

Co-founder of Databricks and Perplexity AI · Founder of Laude Institute
Bayshore Ballroom

10:15 AM 10:45 AM

AM Break

Bayshore Foyer

10:45 AM 12:15 PM

Paper Session 1: Agent Design

Bayshore Ballroom · 9 talks · 7+2+1 min each

10:45 AM	Context, Reasoning, and Hierarchy: A Cost–Performance Study of Compound LLM Agent Design in an Adversarial POMDP
10:55 AM	Expansion-Contraction: A Multi-Agent Graph Traversal Pattern for Compound AI Systems
11:05 AM	A Language for Describing Agentic LLM Contexts
11:15 AM	Tressoir: Unifying Online, Offline, and HIL Design and Evolution of Multi-Agent Systems via Interpretable Blueprints
11:25 AM	Glia: A Human-Inspired AI for Automated Systems Design and Optimization
11:35 AM	Improving Coherence and Persistence in Agentic AI for System Optimization
11:45 AM	Robust Agent Compensation (RAC): Teaching AI Agents to Compensate
11:55 AM	fastWorkflow: Closing the Performance Gap Between Small and Frontier Language Models for Conversational Agents
12:05 PM	TraceFix: Repairing Agent Coordination Protocols with TLA+ Counterexamples

12:15 PM 1:30 PM

Lunch

Gateway Ballroom

1:30 PM 3:00 PM

Paper Session 2: Agent Evaluation

Bayshore Ballroom · 9 talks · 7+2+1 min each

1:30 PM	Trace-Level Analysis of Information Contamination in Multi-Agent Systems
1:40 PM	Willful Disobedience: Automatically Detecting Failures in Agentic Traces
1:50 PM	OpaqueToolsBench: Learning Nuances of Tool Behavior Through Interaction
2:00 PM	ViBench: A Benchmark on Vibe Coding
2:10 PM	Vibe Code Bench: Evaluating AI Models on End-to-End Web Application Development
2:20 PM	Reasoning-Intensive Regression
2:30 PM	Generating Expressive and Customizable Evals for Timeseries Data Analysis Agents with AgentFuel
2:40 PM	Benchmarking Agents in Insurance Underwriting Environments
2:50 PM	DraftNEPABench: A Benchmark for Drafting NEPA Document Sections with Coding Agents

3:00 PM 3:30 PM

PM Break

Bayshore Foyer

3:30 PM 4:30 PM

Paper Session 3: Systems Efficiency

Bayshore Ballroom · 6 talks · 7+2+1 min each

3:30 PM	Robust Batch-Level Query Routing for Large Language Models under Cost and Capacity Constraints
3:40 PM	Understanding and Improving Communication Performance in Multi-node LLM Inference
3:50 PM	Echo: KV-Cache-Free Associative Recall with Spectral Koopman Operators
4:00 PM	XGrammar-2: Dynamic and Efficient Structured Generation Engine for Agentic LLMs
4:10 PM	CAMI: Practical Cost-Aware Agent-Guided Multi-Indexing for Semantic Retrieval
4:20 PM	SwiftFusion: Scalable Sequence Parallelism for Distributed Inference of Diffusion Transformers on GPUs

4:30 PM 5:00 PM

Ops Experience 1: Talk title TBD

Bayshore Ballroom

5:15 PM 6:45 PM

Demos & Research Posters

15 demos · San Jose / Santa Clara
25 research posters · Carmel / Monterey

Demos

Architectural Patterns & Composition12 System Optimization & Efficiency3

Architectural Patterns & Composition 12

System Optimization & Efficiency 3

Browse all accepted demos →

Research Posters

Paper Session 1: Agent Design 9

Paper Session 2: Agent Evaluation 9

Paper Session 3: Systems Efficiency 6

6:00 PM 7:30 PM

Databricks Reception

Sprigs · Capacity-capped at 100 people

Independently Organized Events

The following event takes place at an external venue and is independently organized and hosted by its respective organizers. Listed as a courtesy to attendees and does not constitute sponsorship, endorsement, or official affiliation with ACM CAIS.

SkillsBench 1.0 Launch Party

Evening · 501 2nd St, San Francisco · Organized by BenchFlow, Google DeepMind & Kernel Labs

Afterparty for the Agent Skills'26 workshop with live demos and talks on agent benchmarking. Approval required; RSVP below.

RSVP on Luma