What happened?
How can agents fuse noisy operational signals, reason causally and counterfactually, and localize root causes rather than merely correlate symptoms?
- Grounded diagnosis
- Causal and counterfactual reasoning
- Topology-aware inference
Accepted
A²NetOps 2027 · AAAI-27 Workshop · Montréal
Agents for Autonomous
Network Operations
Causal Reasoning, Uncertainty, and Safe Action
A research forum for AI agents that can diagnose operational failures, decide under uncertainty, and use network-automation tools with safeguards.
FormatOne-day workshopIn-person talks · remote attendance for registered workshop participants
WhenFebruary 22 or 23, 2027Exact day to be assigned
WhereMontréal, CanadaPalais des congrès
Submission deadlineNovember 20, 2026Anywhere on Earth (UTC−12)
01 Overview
Autonomy begins with evidence,
not with an action.
AI agents operating on live networks must do more than answer questions. They must combine topology, telemetry, logs, alarms, tickets, configuration state, and tool outputs; determine what evidence to gather next; distinguish causes from correlations; and know when to act, request approval, roll back, or escalate.
This workshop brings together foundational AI researchers and practitioners who build and operate real networks. We treat network operations as a concrete scientific testbed, network automation as the enabling capability, and autonomous networking as the longer-term objective.
We especially welcome work on building and training domain-specific LLMs and purpose-built models for networking, including domain-specific pretraining, post-training, adaptation, agentic reinforcement learning, and continual, online, federated, or in-network training. We also welcome methods that turn operational experience into safe, persistent, and validated improvement without requiring serving-time model updates.
Who should join
- AI agents
- Causal inference
- Reinforcement learning
- Uncertainty
- Networking
- AIOps
- Domain models
- ML for systems
- Enterprise AI
02 Research questions
Three questions define the workshop.
We focus on the complete path from heterogeneous operational evidence to a verifiable, governed intervention.
What should happen next?
How should an agent gather additional evidence, select tools and APIs, calibrate uncertainty, and choose between acting, asking, and escalating?
- Active evidence gathering
- Sequential decisions
- Risk-aware escalation
How can it act safely?
How can network actions be validated, governed, audited, rolled back, and verified under real operational constraints?
- Safe tool and API use
- Rollback-aware remediation
- Post-action verification
Topics of interest
Work across methods, systems, and evaluation is welcome.
Causal operational diagnosis
Root-cause localization, intervention analysis, and counterfactual reasoning.
Active evidence gathering
Tool selection, stopping rules, partial observability, and delayed outcomes.
Agentic network automation
Intent translation, configuration validation, planning, and closed-loop control.
Domain models, training, and adaptation
Domain-specific pretraining, post-training, agentic reinforcement learning, and continual, online, federated, or in-network training.
Evaluation and validation
Controlled experiments, process-level metrics, reliability testing, and reproducible protocols.
Safe autonomy and oversight
Human approval, governed execution, auditability, rollback, and verification.
03 Invited speakers
Confirmed perspectives from research and deployed AI.
Talk titles and biographies will be announced with the program. Affiliations reflect current information provided by the organizers.
ConfirmedTing He
Professor of Computer Science and Engineering
Pennsylvania State University
ConfirmedJP Vasseur
Senior Distinguished Engineer & Chief Architect of AI and Networking
NVIDIA
ConfirmedJian Tan
Senior Staff Software Engineer · LLM Training
Google
ConfirmedEmily Pitler
Gemini computer use lead
Google DeepMind
ConfirmedAli Dabir
Principal Machine Learning Engineer
Cisco
ConfirmedYury Sokolov
Principal AI Researcher
Cisco04 Program
Designed for exchange, not a day of back-to-back lectures.
The final agenda will follow the session and break times assigned by AAAI. Exact talk titles, times, room, and workshop day remain to be announced.
Shared questions, scope, and goals
AI methods, networks, and deployed systems
Orals selected through peer review
Learning, validation, and empirical findings
A structured community research agenda
Hands-on systems and direct interaction
05 Call for participation
Help define the next generation of autonomous NetOps.
We welcome research, work in progress, well-grounded positions, deployed-system experience, and demonstrations.
OpenReview · opens Oct 16
Contribution types
Contributions should advance reliable, auditable, and controllable agents for network automation and autonomous network operations.
Full papers · up to 8 pagesShort papers · up to 4 pagesPosition papers · up to 4 pagesDemo papers · up to 4 pages
Submission requirements
Full papers may be up to 8 pages; short, position, and demo papers may be up to 4 pages. Page limits include references. Use the two-column format in the AAAI-27 Author Kit (ZIP). Submissions will be evaluated for relevance, technical quality, clarity, and potential to stimulate discussion. At least one author of each accepted submission must present in person. AAAI does not publish workshop papers in its Digital Library; titles, abstracts, and author lists of accepted submissions will be public on OpenReview, while submission PDFs will remain private unless the organizers later announce and explicitly enable their release.
Authors should create or update their OpenReview profiles well before the deadline. New profiles without an institutional email may require moderation; profiles created with an institutional email are activated automatically.
Presentation & remote attendance
All presentations must be delivered in person; virtual presentations are not permitted. Registered workshop participants may attend remotely through AAAI-provided facilities.
Registration
We expect approximately 40–60 participants. Attendance is open to registered AAAI-27 workshop participants on a first-come, first-served basis, subject to room capacity. See AAAI-27 registration details.
OpenReview & publication
Titles, abstracts, and author lists of accepted submissions will be public on OpenReview. Submission PDFs will remain private unless the organizers later announce and explicitly enable their release. AAAI does not publish workshop papers in the AAAI Digital Library.
06 Organizers
Connecting AI, networking, and deployment.
Questions about the workshop may be directed to the workshop chair and main contact.
Workshop chair & main contact
Yuqing Gao
IEEE Fellow & Senior Director, AI Engineering and Research · Cisco
yuqingga@cisco.comWorkshop co-chair
Ting He
Professor of Computer Science and Engineering · Pennsylvania State University
tinghe@psu.eduWorkshop co-chair
JP Vasseur
Senior Distinguished Engineer & Chief Architect of AI and Networking · NVIDIA
jvasseur@nvidia.com
Program chair
Fan BuCisco
fabu@cisco.com
Program committee18 members
- Slava AgafonovCisco
- Hamid AlmasiCisco
- Ali DabirCisco
- Derek EngiCisco
- Lei JinCisco
- Doosan JungCisco
- Ashkan KazemiCisco
- Devavrat KhanolkarCisco
- Rishabh KumarAirbnb
- Ye LiCisco
- Jisheng LiangAirbnb
- Sumeet RamPayPal
- Boris SobolevCisco
- Yury SokolovCisco
- Jian TanGoogle
- Cheng WuMicrosoft
- Ling WuGoogle
- Li ZhangCisco
07 FAQ
What participants should know.
Is this a cybersecurity workshop?
No. The workshop focuses on network operations, network automation, and autonomous networking rather than threat, attack, malware, or defense methods.
Can I present remotely?
No. All presentations must be delivered in person; virtual presentations are not permitted. Registered workshop participants may attend remotely through AAAI-provided facilities.
Will papers appear in the AAAI Digital Library?
No. AAAI does not publish workshop papers in the AAAI Digital Library. Titles, abstracts, and author lists of accepted submissions will be public on OpenReview; submission PDFs will remain private unless the organizers later announce and explicitly enable their release.
Which day is the workshop?
AAAI-27 workshops run February 22–23, 2027. The exact day and room for this one-day workshop will be posted after AAAI assigns them.