pith. sign in

arxiv: 2606.18000 · v1 · pith:WOLB6I7Mnew · submitted 2026-06-16 · 💻 cs.NI · cs.AI

A T-API-Compliant ReAct Agentic Loop for Optical Networks: Generic vs. Domain-Specific Tool Abstractions

Seyed Morteza Ahmadian , Paolo Monti , Carlos Natalino This is my paper

Pith reviewed 2026-06-26 22:10 UTC · model grok-4.3

classification 💻 cs.NI cs.AI
keywords T-APIReAct agentoptical networksdomain-specific toolsintent-driven managementclosed-loop automationagentic systemstool abstractions
0
0 comments X

The pith

The first T-API-compliant ReAct loop for optical networks shows domain-specific composite tools reaching 90 percent oracle-validated correctness with threefold token savings over generic tools.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper builds an agentic management system for optical networks that follows the T-API standard and uses a ReAct loop to reason and act on intents in a closed loop. It directly compares two tool designs: generic abstractions versus domain-specific composite tools. The domain-specific version delivers markedly higher correctness and lower token consumption on validation tasks. A reader would care because optical networks require increasing autonomy to handle dynamic traffic and failures without constant manual oversight.

Core claim

We present the first T-API-compliant reasoning and act (ReAct) loop. We show that domain-specific composite tools achieve 90% oracle-validated correctness with threefold token savings compared to generic tools.

What carries the argument

T-API-compliant ReAct loop that selects between generic tool abstractions and domain-specific composite tools to execute intent-driven closed-loop actions on optical networks.

If this is right

  • Intent-driven closed-loop management becomes feasible at higher autonomy levels for optical networks.
  • Domain-specific composite tools reduce operational token costs while preserving high task correctness.
  • The T-API standard can serve as the interface layer for agentic control loops in production networks.
  • Generic tool sets are less efficient for domain tasks and can be replaced without losing compliance.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same composite-tool pattern could be tested on other standardized interfaces such as OpenConfig or ONF models to check portability.
  • If oracle accuracy holds in live traffic, operators could reduce human-in-the-loop interventions for routine provisioning and recovery.
  • Token savings may translate to lower inference latency and cost when the loop runs on edge or cloud LLM instances.

Load-bearing premise

The oracle used for validation provides an accurate and unbiased measure of real-world correctness, and the experimental comparison between generic and domain-specific tools controls for all relevant confounding factors such as prompt design and task selection.

What would settle it

A side-by-side run of the same ReAct loop on a live optical network testbed where domain-specific tools fall below 80 percent end-to-end success rate or show no measurable token reduction.

Figures

Figures reproduced from arXiv: 2606.18000 by Carlos Natalino, Paolo Monti, Seyed Morteza Ahmadian.

Figure 1
Figure 1. Figure 1: The T-API-Compliant ReAct Agentic Architecture. [7, 8, 16] and plan-and-execute prior work [6, 20], our loop iteratively reasons, calls a tool, observes, and decides dynamically, according to instructions. Two tool alternatives share the same agent and the same control and data planes. The first al￾ternative employs generic HTTP tools, that is, two primitives (GET/POST with URL and body) following the phil… view at source ↗
Figure 2
Figure 2. Figure 2: Evaluation pass rate (a) and mean total tokens per run (b) for different LLM models and tool abstractions. ness, combining tool-presence checks with value checks against ground-truth quantities fetched from the DT. The oracle further types each failure as zero-tool, wrong-value, wrong-modulation, or missing-grounding, thereby extending the quali￾tative taxonomies of [13] and [21]. Provisioning is scored in… view at source ↗
read the original abstract

Optical networks need intent-driven, closed-loop agentic management, a key enabler for higher autonomy levels. We present the first T-API-compliant reasoning and act (ReAct) loop. We show that domain-specific composite tools achieve 90% oracle-validated correctness with threefold token savings compared to generic tools.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript presents the first T-API-compliant ReAct agentic loop for intent-driven optical network management. It claims that domain-specific composite tools achieve 90% oracle-validated correctness with threefold token savings relative to generic tools.

Significance. If the empirical comparison holds after proper controls and oracle definition, the result would supply concrete evidence favoring domain-specific abstractions in agentic telecom systems and could inform tool design for higher autonomy levels.

major comments (1)
  1. The headline quantitative claims (90% oracle-validated correctness and 3× token savings) are load-bearing for the preference of domain-specific tools, yet the manuscript provides no description of the oracle, the task dataset, prompt controls, baseline tool implementations, or error analysis. Without these, it is impossible to determine whether the oracle is independent or whether task selection and prompt wording confound the generic vs. domain-specific comparison.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on the experimental details supporting our quantitative claims. We agree these elements are necessary for rigorous evaluation and will expand the manuscript accordingly.

read point-by-point responses

  1. Referee: The headline quantitative claims (90% oracle-validated correctness and 3× token savings) are load-bearing for the preference of domain-specific tools, yet the manuscript provides no description of the oracle, the task dataset, prompt controls, baseline tool implementations, or error analysis. Without these, it is impossible to determine whether the oracle is independent or whether task selection and prompt wording confound the generic vs. domain-specific comparison.

    Authors: We acknowledge that the current manuscript provides insufficient detail on these elements. In the revised version we will add a new subsection in the Experiments section that explicitly defines: the oracle as an independent rule-based validator derived from T-API specifications and cross-validated by two domain experts; the task dataset of 50 intent scenarios across three topologies; prompt controls consisting of fixed system prompts, temperature=0, and identical few-shot examples (or none) for both conditions; baseline implementations with pseudocode for generic per-call tools versus the domain-specific composites; and a full error analysis breaking down failures by category (reasoning, tool invocation, oracle mismatch). These additions will permit assessment of independence and potential confounds. The reported aggregate results will be re-checked against the expanded documentation but are not expected to change. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical system presentation with no derivation chain

full rationale

The paper describes an empirical implementation and evaluation of a T-API-compliant ReAct agentic loop for optical networks, reporting observed performance differences (90% oracle-validated correctness and threefold token savings) between domain-specific composite tools and generic tools. No equations, fitted parameters, predictions derived from first principles, or load-bearing self-citations appear in the abstract or described content. The central claims rest on experimental observations rather than any reduction of outputs to inputs by construction, self-definition, or imported uniqueness theorems. The work is therefore self-contained as an empirical contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no information on free parameters, background axioms, or newly postulated entities.

pith-pipeline@v0.9.1-grok · 5576 in / 969 out tokens · 33059 ms · 2026-06-26T22:10:39.430708+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

24 extracted references · 11 canonical work pages · 1 internal anchor

  1. [1]

    Autonomous networks: In search of best practice

    R. Webb and D. Bushaus, “Autonomous networks: In search of best practice”, TM Forum, Tech. Rep., Dec. 2024, p. 51

    2024

  2. [2]

    Framework and overall objectives of the future development of IMT for 2030 and beyond

    ITU-R, “Framework and overall objectives of the future development of IMT for 2030 and beyond”, International Telecommunication Union, Geneva, Switzerland, Rec- ommendation ITU-R M.2160-0, Nov. 2023

    2030

  3. [3]

    Zero-touch network and service man- agement (ZSM); intent-driven closed loops

    ETSI ISG ZSM, “Zero-touch network and service man- agement (ZSM); intent-driven closed loops”, European Telecommunications Standards Institute, Sophia Antipo- lis, France, Group Specification ETSI GS ZSM 016 V1.1.1, Oct. 2024

    2024

  4. [4]

    The role of digital twin in optical com- munication: Fault management, hardware configuration, and transmission simulation

    D. Wang et al., “The role of digital twin in optical com- munication: Fault management, hardware configuration, and transmission simulation”,IEEE Communications Magazine, vol. 59, no. 1, pp. 133–139, Jan. 2021.DOI: 10.1109/MCOM.001.2000727

    work page doi:10.1109/mcom.001.2000727 2021

  5. [5]

    First field trial of llm-powered ai agent for lifecycle management of autonomous driving optical networks

    X. Liu et al., “First field trial of llm-powered ai agent for lifecycle management of autonomous driving optical networks”, in2025 Optical Fiber Communications Con- ference and Exhibition (OFC), Mar. 2025, Th1A.2.DOI: 10.1364/OFC.2025.Th1A.2

    work page doi:10.1364/ofc.2025.th1a.2 2025

  6. [6]

    Field trial of an llm-powered AI agent for autonomous optical networks: A full-lifecycle demonstration

    X. Liu, Y . Zhang, Q. Qiu, Y . Cheng, W. Hu, and Q. Zhuge, “Field trial of an llm-powered ai agent for autonomous optical networks: A full-lifecycle demonstration”,Jour- nal of Optical Communications and Networking, vol. 18, no. 5, A179, May 2026.DOI:10.1364/JOCN.575402

    work page doi:10.1364/jocn.575402 2026

  7. [7]

    Multi-agent design for llm-assisted network management

    H. Zaid, P . Safari, B. Shariati, A. Jafari, M. Balanici, and J. K. Fischer, “Multi-agent design for llm-assisted network management”, in2025 Optical Fiber Commu- nications Conference and Exhibition (OFC), Mar. 2025, W1A.4.DOI:10.1364/OFC.2025.W1A.4

    work page doi:10.1364/ofc.2025.w1a.4 2025

  8. [8]

    Data sovereign llm-assisted automa- tion platform for open optical and packet transport networks

    B. Shariati et al., “Data sovereign llm-assisted automa- tion platform for open optical and packet transport networks”, in2025 IEEE International Conference on Machine Learning for Communication and Network- ing (ICMLCN), May 2025, pp. 1–6.DOI: 10 . 1109 / ICMLCN64995.2025.11140539

    arXiv 2025

  9. [9]

    Large language model-driven ai agent in sdn controller towards intent-based management of optical networks

    A. Zhou, Y . Song, Y . Zhang, M. Zhang, and D. Wang, “Large language model-driven ai agent in sdn controller towards intent-based management of optical networks”, inECOC 2024; 50th European Conference on Optical Communication, Sep. 2024, pp. 1595–1598

    2024

  10. [10]

    Field trial of llm-based autonomous net- work management with ai-agent in real-time 400g/800g elastic optical network

    H. Huang et al., “Field trial of llm-based autonomous net- work management with ai-agent in real-time 400g/800g elastic optical network”, in2025 European Confer- ence on Optical Communications (ECOC), Sep. 2025, p. M.03.01.3.DOI: 10 . 1109 / ECOC66593 . 2025 . 11263349

    2025

  11. [11]

    Cross-domain orchestration with multi- agent llm framework for enhanced task automation

    X. Xu et al., “Cross-domain orchestration with multi- agent llm framework for enhanced task automation”, in 2025 Optical Fiber Communications Conference and Exhibition (OFC), Mar. 2025, M3Z.10.DOI: 10.1364/ OFC.2025.M3Z.10

    2025

  12. [12]

    Model human learners: Com- putational models to guide instructional design.arXiv preprint arXiv:2502.02456, 2025

    Z. Wang et al.,Agentic ai for scalable and robust opti- cal systems control, Feb. 2026.DOI: 10.48550/arXiv. 2602.20144arXiv:2602.20144 [eess]

    work page internal anchor Pith review doi:10.48550/arxiv 2026

  13. [13]

    Opticomm-gpt: A gpt-based versatile research assistant for optical fiber communication sys- tems

    X. Jiang et al., “Opticomm-gpt: A gpt-based versatile research assistant for optical fiber communication sys- tems”,Optics Express, vol. 32, no. 12, pp. 20 776– 20 796, Jun. 2024.DOI:10.1364/OE.522026

    work page doi:10.1364/oe.522026 2024

  14. [14]

    On-premises small lan- guage model agent with physics-aware reasoning for optical network optimization

    T. Tanimura and N. Kikuchi, “On-premises small lan- guage model agent with physics-aware reasoning for optical network optimization”, inOptical Fiber Communi- cation Conference (OFC), 2026, Th2A.33

    2026

  15. [15]

    LLM-enhanced digital twin framework in optical networks

    S. Shen et al., “LLM-enhanced digital twin framework in optical networks”, inOptical Fiber Communication Conference (OFC), 2026, M4A.6

    2026

  16. [16]

    The case for a dnanf 1 pb/s trans-atlantic submarine cable

    A. Jafari et al., “Llm assistant for tapi context and client code translation”, in2025 European Conference on Op- tical Communications (ECOC), Sep. 2025, p. M.03.01.2. DOI:10.1109/ECOC66593.2025.11263212

    work page doi:10.1109/ecoc66593.2025.11263212 2025

  17. [17]

    ACM Netw.2, DOI: 10.1145/3656296 (2024)

    C. Wang, M. Scazzariello, A. Farshin, S. Ferlin, D. Kosti´c, and M. Chiesa, “NetConfEval: Can LLMs facilitate net- work configuration?”,Proceedings of the ACM on Net- working, vol. 2, no. CoNEXT2, 7:1–7:25, Jun. 2024.DOI: 10.1145/3656296

    work page doi:10.1145/3656296 2024

  18. [18]

    Toolllm: Facilitating large language models to master 16000+ real-world apis

    Y . Qin et al., “Toolllm: Facilitating large language models to master 16000+ real-world apis”, inInternational Con- ference on Learning Representations, vol. 2024, May 2024, pp. 9695–9717

    2024

  19. [19]

    Autoonbench: A benchmark for large language model agents in autonomous optical networks

    Y . Zhang, Q. Qiu, J. Wu, X. Liu, W. Hu, and Q. Zhuge, “Autoonbench: A benchmark for large language model agents in autonomous optical networks”,Journal of Op- tical Communications and Networking, vol. 18, no. 9, pp. D1–D15, Sep. 2026.DOI:10.1364/JOCN.589201

    work page doi:10.1364/jocn.589201 2026

  20. [20]

    When large language models meet opti- cal networks: Paving the way for automation

    D. Wang, Y . Wang, X. Jiang, Y . Zhang, Y . Pang, and M. Zhang, “When large language models meet opti- cal networks: Paving the way for automation”,Elec- tronics, vol. 13, no. 13, Jun. 2024.DOI: 10 . 3390 / electronics13132529

    2024

  21. [21]

    Enhancing large language models (LLMs) for telecom using dynamic knowledge graphs and explain- able retrieval-augmented generation

    D. Yuan, H. Zhou, X. Liu, H. Chen, Y . Xin, and J. Zhang, “Enhancing large language models (LLMs) for telecom using dynamic knowledge graphs and explain- able retrieval-augmented generation”,IEEE Wireless Communications, 2026, Accepted; arXiv:2602.17529

    arXiv 2026

  22. [22]

    Applicability of MCP for the network management

    Y . Y ang, Q. Wu, D. Lopez, N. R. Moreno, and L. Tailhar- dat, “Applicability of MCP for the network management”, Internet Engineering Task Force, Internet Draft draft- yang-nmrg-mcp-nm-02, Feb. 2026

    2026

  23. [23]

    Gnpy model of the physical layer for open and disaggregated optical networking [invited]

    V. Curri, “Gnpy model of the physical layer for open and disaggregated optical networking [invited]”,Journal of Optical Communications and Networking, vol. 14, no. 6, pp. C92–C104, Jun. 2022.DOI:10.1364/JOCN.452868

    work page doi:10.1364/jocn.452868 2022

  24. [24]

    Optical networking gym: An open- source toolkit for resource assignment problems in opti- cal networks

    C. Natalino et al., “Optical networking gym: An open- source toolkit for resource assignment problems in opti- cal networks”,Journal of Optical Communications and Networking, vol. 16, no. 12, G40–G51, Dec. 2024.DOI: 10.1364/JOCN.532850

    work page doi:10.1364/jocn.532850 2024