WG Working Group R. Schott Internet-Draft Deutsche Telekom Intended status: Informational J. Maisonneuve Expires: 31 August 2026 Nokia Bell Labs L. M. Contreras Telefonica J. Ros-Giralt Qualcomm Europe, Inc. 27 February 2026 Agentic AI Use Cases draft-scrm-aiproto-usecases-latest Abstract TODO Abstract About This Document This note is to be removed before publishing as an RFC. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-scrm-aiproto-usecases/. Source for this draft and an issue tracker can be found at https://github.com/https://github.com/giralt/draft-scrm-aiproto- usecases. Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at https://datatracker.ietf.org/drafts/current/. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on 31 August 2026. Copyright Notice Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/ license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Table of Contents 1. Introduction 2. Conventions and Definitions 3. Use Cases 3.1. Deep Search 3.1.1. Building Blocks 3.1.2. Why this use case matters 3.1.3. Example: Open Deep Search project 3.1.4. Informative References 4. Security Considerations 5. IANA Considerations 6. Normative References Acknowledgments Authors' Addresses 1. Introduction TODO Introduction 2. Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here. 3. Use Cases Roland use case test edition ... 3.1. Deep Search *Deep Search* refers to an _agentic_ information-seeking workflow in which an AI agent plans, executes, and iteratively refines multi-step research across heterogeneous sources (open web, enterprise knowledge bases, APIs, files, and computational tools). Unlike one-shot retrieval or a single RAG call, Deep Search is *long-horizon* and *goal-directed*: the agent decomposes a task into sub-goals, issues searches and crawls, reads and filters evidence, runs auxiliary computations (e.g., code or math), verifies claims, tracks provenance/citations, and synthesizes a final answer—often over minutes or hours rather than milliseconds. This loop is typically implemented as _think → act (tool) → observe → reflect → refine plan_ until success criteria (coverage, confidence, cost/time budgets) are met. 3.1.1. Building Blocks A conformant Deep Search workflow comprises the following components: * *Base LLM (reasoning + tool use).* A model capable of multi-step reasoning (e.g., chain-of-thought/verification, self-reflection) _and_ structured tool invocation (function/tool calling) to plan, call tools, parse results, and revise plans. * *Planner/Orchestrator.* A lightweight controller (can be the LLM itself or a graph/agent runtime) that maintains task state, goals, and budgets (tokens, latency, rate limits), and schedules steps (parallel vs. sequential). * *Tooling layer (invoked by the agent):* - *Web search & crawling* (SERP APIs, focused crawlers, HTML/PDF parsers, robots.txt compliance). - *Knowledge access* (enterprise KBs, document stores, wikis, code repos). - *Retrieval-Augmented Generation (RAG)* (indexing, re-ranking, query rewriting, dedup, chunking). - *Computation tools* (e.g., *Python interpreter* for factual checks, data wrangling, statistics/plots). - *Specialized services* (scholarly search, calculators, geocoders, OCR, table extraction, etc.). - *Verification/critique* (fact-checking, citation validation, deduplication, hallucination detection). - *Provenance & citation store* (source URIs, timestamps, quotes/ snippets, hashes). * *Short-term memory / working set.* Scratchpad to hold the evolving evidence graph: normalized documents, extracted entities/claims, metadata, and confidence scores. * *Synthesis & reporting.* Templates or renderers that compile the final artifact (report/brief/bibliography), with explicit citations to the evidence used. * *Observability & policy.* Logging, traces, and red-team hooks for auditability; safety filters (PII, ToS, copyright/robots policy), rate limiting, attribution. +--------------------------------------+ | Planning Loop | | (reason over goals, budgets, time) | +--------------------------------------+ | (1) User Goal / Query | v +------------------+ plan / subgoals +------------------------+ | Reasoning LLM |<----------------------->| Orchestrator / | | (tool-capable) | reflect / revise plan | Task Controller | +------------------+ +------------------------+ | | | (2) tool calls | (3) schedule/monitor v v +------------------+ +------------------+ +------------------+ +------------------+ | Web Search / | | Crawlers | | KB / RAG Index | | Python / Tools | | SERP APIs |--->| (focused fetch) |--->| (embed/rerank) |--->| (compute, eval) | +------------------+ +------------------+ +------------------+ +------------------+ \___________________________ | _______________________________________________/ v +-----------------------------+ | Evidence / Working Set | | (docs, snippets, claims, | | citations, confidence) | +-----------------------------+ | (4) verify / critique v +---------------------+ | Verifier/Critic | | (consistency, | | provenance checks) | +---------------------+ | (5) synthesize v +--------------------------------------+ | Report / Answer with Citations | | (export, stream, or hand off) | +--------------------------------------+ Figure 1: Deep Search agentic workflow The loop repeats until success criteria are met (coverage/quality thresholds, budget, or explicit user stop). 3.1.2. Why this use case matters Deep Search is inherently *compositional*: it coordinates _multiple_ agents and _many_ tools over extended time. Without standard protocols, systems devolve into brittle, one-off integrations that are hard to test, secure, or reuse. Two complementary interoperability layers are especially relevant: * *Agent-to-Tool standardization (inside an agent).* The *Model Context Protocol (MCP)* defines a common way for agents/hosts to discover, describe, and invoke tools, resources, and prompts over JSON-RPC across transports (stdio, HTTP/SSE, WebSocket). MCP enables portable tool catalogs (search, crawler, RAG, Python) with consistent schemas, capability negotiation, progress/cancellation, and security prompts/consent. Model Context Protocol specification (https://modelcontextprotocol.io/ specification/2025-03-26) • MCP GitHub org (https://github.com/ modelcontextprotocol). * *Agent-to-Agent standardization (between agents/systems).* The *Agent2Agent (A2A)* protocol focuses on inter-agent collaboration—capability discovery (Agent Cards), task lifecycle (create/cancel/status), streaming updates for long-running jobs, and opaque collaboration without sharing proprietary internals. A2A complements MCP (A2A connects _agents_; MCP connects an agent to its _tools_). See the overview/spec and announcement: A2A protocol site/spec (https://a2a-protocol.org/latest/) • Google Developers announcement (https://developers.googleblog.com/en/a2a- a-new-era-of-agent-interoperability/) • A2A GitHub (https://github.com/a2aproject/A2A). The IETF has also created a non-WG *agent2agent* list to scope standardization work in this space (IETF mail archive (https://mailarchive.ietf.org/arch/msg/agent2agent/ a6ORMj_pMlYZI_KU3gPMhB-K9Ng/)). *Implications for Deep Search.* Using A2A and MCP together lets implementers compose portable Deep Search stacks: * Tools like crawlers, scholarly search, RAG, and Python are exposed via *MCP* with typed inputs/outputs and consent flows. * Long-running research tasks, delegation to specialized researcher/ verifier agents, background execution, progress streaming, and result handoff occur via *A2A*. * Provenance (URIs, hashes, timestamps) and citation schemas can be standardized at the protocol boundary to enable verifiable research traces across vendors. * Enterprise requirements—authn/z, quotas, observability/tracing, policy enforcement (robots/copyright), and safety reviews—become portable rather than per-integration glue. 3.1.3. Example: Open Deep Search project Open implementations illustrate agentic architectures for Deep Search: * *Open Deep Search (ODS).* A modular open-source framework that augments a base LLM with a *Reasoning Agent* and an *Open Search Tool*, reporting state-of-the-art results on benchmarks like SimpleQA and FRAMES; paper and code are available. Alzubi et al., 2025 (arXiv) (https://arxiv.org/abs/2503.20201) • sentient-agi/ OpenDeepSearch (GitHub) (https://github.com/sentient-agi/ OpenDeepSearch). * *Open Deep Research (LangChain).* An agentic “deep research” reference built on LangGraph that works across model providers, search tools, and *MCP servers*; includes supervisor/sub-agent patterns and evaluation harnesses. Project blog (https://www.blog.langchain.com/open-deep-research/) • langchain-ai/open_deep_research (GitHub) (https://github.com/ langchain-ai/open_deep_research). These systems exemplify the building blocks in §2 and are consistent with the interoperability layering in §4 (MCP for tools; A2A for inter-agent collaboration). 3.1.4. Informative References * Model Context Protocol (spec & docs): modelcontextprotocol.io (https://modelcontextprotocol.io/specification/2025-03-26), GitHub organization (https://github.com/modelcontextprotocol). * Agent2Agent Protocol (A2A): a2a-protocol.org (https://a2a- protocol.org/latest/), spec (https://a2a-protocol.org/latest/ specification/), Google Developers announcement (https://developers.googleblog.com/en/a2a-a-new-era-of-agent- interoperability/), Linux Foundation news (https://developers.googleblog.com/en/google-cloud-donates-a2a-to- linux-foundation/). * IETF “agent2agent” list (standardization discussion): mailarchive.ietf.org (https://mailarchive.ietf.org/arch/msg/agent2agent/ a6ORMj_pMlYZI_KU3gPMhB-K9Ng/). * Open Deep Search (ODS): arXiv:2503.20201 (https://arxiv.org/ abs/2503.20201), GitHub repository (https://github.com/sentient- agi/OpenDeepSearch). * LangChain Open Deep Research: project blog (https://www.blog.langchain.com/open-deep-research/), GitHub (https://github.com/langchain-ai/open_deep_research). `` 4. Security Considerations TODO Security 5. IANA Considerations This document has no IANA actions. 6. Normative References [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, . [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, . Acknowledgments TODO acknowledge. Authors' Addresses Roland Schott Deutsche Telekom Email: Roland.Schott@telekom.de Julien Maisonneuve Nokia Bell Labs Email: julien.maisonneuve@nokia.com L. M. Contreras Telefonica Email: luismiguel.contrerasmurillo@telefonica.com Jordi Ros-Giralt Qualcomm Europe, Inc. Email: jros@qti.qualcomm.com