The Future is Agentic in Recommender Systems

2026-04-25 · Source: Data Skeptic · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, extended

Summary

Yashar Deldjoo, an Associate Professor at the Polytechnic University of Bari and a senior research scientist, discusses the evolution of recommender systems, emphasizing the growing importance of trustworthiness and the impact of Large Language Models (LLMs). The conversation covers key dimensions of responsible AI, including generalizability, robustness against adversarial attacks, privacy, explainability, and fairness. Deldjoo highlights new risks introduced by LLMs, such as hallucinations and context drift, which can lead to inaccurate or irrelevant recommendations. He also explores "agentic" recommender systems, where LLMs act as a central "brain" using various tools and memory types to move beyond simple ranked lists to accomplish complex, multi-constraint tasks, like personalized travel planning. Deldjoo's forthcoming book, "Recommendations with Generative Models," further details these concepts, categorizing generative models by data modality and addressing their evaluation and associated ethical risks.

Key takeaway

For AI Architects and Product Managers designing next-generation recommendation platforms, you should prioritize integrating robust trustworthiness frameworks from the outset. While LLMs offer unprecedented capabilities for conversational and agentic recommendations, their inherent risks like hallucination necessitate careful alignment training and multi-layered safety protocols. Focus on hybrid models that augment proven collaborative filtering with LLM-powered agents to balance innovation with reliability, ensuring your systems can handle complex user tasks without compromising accuracy or safety.

Key insights

LLMs are transforming recommender systems from ranked lists to agentic, task-oriented solutions, while introducing new trustworthiness challenges.

Principles

• Trustworthy AI dimensions include robustness, privacy, explainability, and fairness.
• LLMs introduce new risks like hallucination and context drift.
• Agentic systems use LLMs as a "brain" with tools and memory for complex tasks.

Method

Agentic recommender systems leverage LLMs, external tools, and various memory types (working, episodic, semantic, procedural) to process complex user constraints and execute multi-step tasks, moving beyond traditional ranking to end-to-end task completion.

In practice

• Use LLMs for conversational AI in recommender systems.
• Augment existing recommender systems with LLMs for data generation.
• Simulate user behavior and system responses with LLM agents.

Topics

• Recommender Systems
• Large Language Models
• Agentic Systems
• Trustworthy AI
• Conversational AI

Best for: AI Architect, AI Engineer, AI Product Manager, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• A Foundation Model for Generative Recommendation, A Unified Definition of Hallucination, and More! — Recent research advances generative recommendation, LLM integration, and efficiency in information retrieval systems.
• Trustworthy Agent Network: Trust in Agent Networks Must Be Baked In, Not Bolted On — Trust in LLM-based agent networks must be architected into the system's core design, not retrofitted.
• The Truth About AI Agents That Nobody Wants to Admit — Based on over 1,500 upvotes on Moltbook, AI agents are identified as the paramount topic in AI, signifying a major paradigm shift from "ask and answer" to "tell and do" by enabling autonomous action.
• 2025 Year in Review — Integrating LLMs with RecSys can yield powerful, steerable recommendation systems capable of cross-platform suggestions.
• Agentic Recommender System with Hierarchical Belief-State Memory — Hierarchical memory and adaptive lifecycle management significantly enhance agentic recommender system performance.
• LLM-Enhanced Messaging: The Rise of Agentic Inboxes — LLMs are transforming messaging from simple filtering to agentic inboxes, significantly boosting knowledge worker productivity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Data Skeptic.