[AI Digest] Multi-Agent Systems Advance Customer Experience

Daily AI Research Update - November 15, 2025

Today's AI research landscape reveals groundbreaking advances in multi-agent systems, dialogue management, and tool-augmented reasoning. These developments are reshaping how AI agents coordinate, communicate, and deliver sophisticated customer experiences. From novel planning architectures that optimize complex tool interactions to scalable dialogue systems with advanced memory capabilities, the research community is addressing critical challenges in building next-generation customer experience platforms.

📌 Beyond ReAct: A Planner-Centric Framework for Complex Tool-Augmented LLM Reasoning

Description: This paper introduces a novel Planner-centric Plan-Execute paradigm that addresses limitations in current tool-augmented LLMs. The framework uses global Directed Acyclic Graph (DAG) planning for complex queries, enabling optimized execution beyond conventional tool coordination. It includes a two-stage training methodology combining Supervised Fine-Tuning with Group Relative Policy Optimization.

Category: Web agents

Why it matters: This framework is highly relevant for Anyreach as it solves critical challenges in coordinating multiple tools and services within AI agents. The DAG-based planning approach could significantly improve how web agents handle complex customer queries that require multiple API calls or tool interactions, reducing errors and improving response quality.

Read the paper →

📌 Fixed-Persona SLMs with Modular Memory: Scalable NPC Dialogue on Consumer Hardware

Description: This research proposes a modular dialogue system using Small Language Models (SLMs) with runtime-swappable memory modules. The system maintains character-specific conversational context and world knowledge without retraining, enabling expressive interactions and long-term memory on consumer-grade hardware.

Category: Chat agents

Why it matters: The modular memory architecture and persona-driven approach are directly applicable to customer service chatbots. This could enable Anyreach to deploy more personalized, context-aware chat agents that maintain conversation history efficiently while running on standard hardware, reducing infrastructure costs.

Read the paper →

📌 SlideBot: A Multi-Agent Framework for Generating Informative, Reliable, Multi-Modal Presentations

Description: SlideBot introduces a modular, multi-agent framework that integrates LLMs with retrieval, structured planning, and code generation. The system uses specialized agents that collaboratively retrieve information, summarize content, generate figures, and format outputs, incorporating evidence-based instructional design principles.

Category: Web agents

Why it matters: The multi-agent collaboration approach and integration of retrieval with content generation is valuable for creating web agents that can produce rich, multi-modal responses to customer queries. This could enhance Anyreach's ability to provide comprehensive, visually-enhanced customer support through web interfaces.

Read the paper →

📌 Echoing: Identity Failures when LLM Agents Talk to Each Other

Description: This paper investigates identity failures that occur when LLM agents interact with each other, revealing important considerations for multi-agent systems.

Category: Chat agents

Why it matters: Understanding identity preservation in agent-to-agent communication is crucial for Anyreach when implementing handoffs between different AI agents or when multiple agents need to collaborate on complex customer issues. This research could help prevent confusion and maintain consistent customer experience across agent interactions.

Read the paper →

This research roundup supports Anyreach's mission to build emotionally intelligent, visually capable, and memory-aware AI agents for the future of customer experience.