REALM @ EMNLP 2026

Autonomous agents are rapidly moving from research ideas to real-world applications. OpenClaw and Cowork are recent examples that show how agents have started to enter everyday use.

This rapid progress in agents has sparked a wave of research across diverse directions: tool use (Toolformer; τ-bench; LiveMCPBench), interacting with user interfaces (OSWorld; OpenCUA; WebArena; WorkArena), external or private knowledge sources (RAG; RAS; Interact-RAG; Agentic RAG), memory and world models (MemAgent; World-in-World), planning (ReAct; Reflexion; Hinter), self-improvement (Self-Refine; TT-SI; EvolveR), and safety and security (RedTeaming; Data Leakage; AgentDojo; RedTeamCUA; ARTEMIS; HITL Security). Together, these advances allow agents to tackle tasks that would have seemed out of reach just a few years ago.

Yet, this level of complexity introduces a broader challenge: turning promising agent capabilities into dependable systems. Beyond scaling, agents must become more reliable on long-horizon tasks, be evaluated in realistic settings, integrate planning, memory, and tools effectively, and interact safely with external environments.

In light of these challenges and opportunities, our workshop aims to bring the ACL community together to analyze the current landscape of LLM agents and its key challenges, while envisioning future directions. The success of the first edition edition at ACL 2025 motivated us to continue the workshop at EMNLP 2026.

Venue

The REALM workshop will take place at the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026, October 24–29), in Budapest, Hungary 🇭🇺.

The workshop will be held on Thursday, October 29, from 9:00am to 5:30pm in Room P1, in a hybrid format (in-person and online).

Call For Papers

We invite submissions on research at the intersection of AI agents and LLMs. See the full Call for Papers for topics, submission guidelines, and important dates. Interested in reviewing? See Call for Reviewers on the same page.