Representations and protocols for networked agents
Algorithms for representing and discovering agent capabilities at web scale — capability embeddings, learned network topologies, and retrieval for discovery and resolution.
1st NeurIPS Workshop on Agentic Web
Decentralized, Continually-Adapting Agent Ecosystems
“What new science do we need for agentic intelligence at internet scale?”
About the Workshop
AI's basic unit of computation is shifting — from single models and individual agents toward continually running, internet-scale collectives of agents that discover one another, communicate, coordinate, and adapt over open networks. Studying these collectives raises questions that single-agent methods do not answer: how to represent and discover agent capabilities at web scale; how learning, credit assignment, and self-improvement behave over changing agent graphs; and how large populations of agents coordinate, remain trustworthy, and stay safe under strategic pressure.
This workshop treats the networked agent collective as a first-class research object, convening researchers across multi-agent and representation learning, learning theory, reinforcement learning, and the societal impacts of AI — to identify the field's foundational problems and the shared evaluation it needs. The open problems are fundamentally about learning and adaptation rather than infrastructure, which is why NeurIPS — rather than a systems or protocols venue — is the right home for this discussion.
The workshop will be held in Atlanta, Georgia, co-located with NeurIPS 2026.
Evaluation
Shared benchmarks for the adaptation, stability, and emergent capability of always-on collectives, where no standard exists today.
Learning Dynamics
Credit assignment over changing agent graphs, and when networked self-improvement converges rather than destabilizes.
Trust & Coordination
Manipulation-resistant reputation and stable coordination under strategic agents at scale.
Call for Papers
We invite submissions that treat the network — rather than a single model or agent — as the unit of analysis. We welcome both early-stage work and more mature results, including position pieces, negative results, and datasets or benchmarks. The program spans three coupled research thrusts.
Learning, adaptation, and self-improvement at the network level
Continual and online learning over changing agent graphs; credit assignment across the network; emergent collective intelligence; and holistic evaluation of self-improving systems.
Coordination, governance, and safety
Manipulation-resistant reputation and trust; coordination that aligns large agent populations; population-scale dynamics and emergent norms; and alignment, accountability, and safety for continually running, self-modifying systems.
Submission Guidelines
Short papers
Up to 4 pages
Excluding references and appendices. Ideal for early-stage work, position pieces, and negative results.
Long papers
Up to 9 pages
Excluding references and appendices. For more mature results, datasets, and benchmarks.
- Non-archival. Accepted papers will be available on OpenReview and are non-archival, so they may be submitted elsewhere. Accepted titles will also be posted on this website before the schedule is published.
- Format. Submissions must use the NeurIPS template and are reviewed double-blind through OpenReview.
- Scope. Position pieces, negative results, and datasets or benchmarks are welcome alongside standard research contributions.
Important Dates
-
August 29, 2026Submission deadline
-
September 2–22, 2026Reviewing period
-
September 29, 2026Author notification
-
November 13, 2026Camera-ready & schedule posted
-
December 2026Workshop — exact date TBD, co-located with NeurIPS 2026 in Atlanta, GA
All deadlines are 23:59 anywhere on Earth (AoE). Dates are subject to change.
Schedule
Tentative schedule.
| Time | Session |
|---|---|
| 00:00 – 00:15 | Opening: the major transition to networked agents |
| 00:15 – 01:00 | Keynote 1 |
| 01:00 – 01:45 | Contributed talks (3 × 15 min) |
| 01:45 – 02:15 | Coffee & poster session I |
| 02:15 – 03:00 | Invited talks — block I (foundations and learning) |
| 03:00 – 03:45 | Hands-on demo of state-of-the-art agentic web tools |
| 03:45 – 05:00 | Lunch |
| 05:00 – 05:45 | Keynote 2 |
| 05:45 – 06:30 | Invited talks — block II (economics, governance, safety) |
| 06:30 – 07:00 | Coffee & poster session II |
| 07:00 – 07:45 | Contributed talks (3 × 15 min) |
| 07:45 – 08:30 | Moderated panel & open debate: what is the unit of intelligence? |
| 08:30 – 08:45 | Closing: research agenda and artifact commitments |
Organizers
Pradyumna Chari
MIT
Eric Horvitz
Microsoft
Zixuan Ke
Salesforce AI Research
Risto Miikkulainen
UT Austin & Cognizant AI Lab
Invited speakers and the program committee will be announced soon.
Contact
For questions about the workshop, submissions, or participation, please reach out to the organizers at pchari@media.mit.edu or risto@cs.utexas.edu. We are happy to hear from you.