The slides
Presentation — 16 slides + 2 backup
Self-contained Reveal.js deck that runs in any browser, also offline. Navigate with the arrow keys
or a presenter remote; press S for speaker notes, F for fullscreen.
The talk
Agentic AI is entering network management, and the same task can be handed to an agent through very different interfaces, from raw device commands to high-level intents. This talk argues that the abstraction level of the Agent–Network Interface is a design variable in its own right, and shows how to measure its effect.
- The question. At which level of abstraction should the agent act on the network?
- The playground. ANI-Gamut exposes the same network through a gamut of interfaces, with the abstraction level as a controlled variable.
- The pilot. Reliability and cost follow the abstraction level; collateral safety does not.
Paper & references
- Paper. L. Bracciale, P. Loreti, A. Mayer, S. Salsano, W. Henderickx, “ANI-Gamut: Benchmarking Agent Reliability across the Gamut of Agent–Network Interface Abstractions”, CNSM 2026.
- Extended version. PDF (v2, 28 October 2026) · arXiv:2609.22723
- Code. github.com/netgroup/ani-gamut (tag
cnsm2026) - Project page. netgroup.github.io/ani-gamut
- Earlier work. S. Salsano, A. Mayer, L. Bracciale, P. Loreti, G. Bianchi, “A Playground for Benchmarking Agentic AI in Network Management”, MAIN 2026.
Notes
This work was funded by the Italian MUR PRIN 2022 NEWTON project, co-funded by PNRR / NextGenerationEU (Italia Domani).
Contact: stefano.salsano@uniroma2.it