[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-study-maps-how-llm-agents-should-talk-to-users":10},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":22,"tags":30,"sources":34,"feedback":38,"feedback_at":22,"cost_usd":38,"total_tokens":38},1196,"new-study-maps-how-llm-agents-should-talk-to-users","New study maps how LLM agents should talk to users","Researchers propose a Communication Policy framework and an evolution method that boost task success without changing the underlying model.","LLMs need smarter ways to talk to people, and a new paper sketches how.\n\nThe authors formalize a “Communication Policy” that defines what an autonomous LLM agent should say and how it should use UI elements. They test text‑only, UI‑only, and hybrid approaches across several simulated tasks, personas, and model pairings. Results show text excels at raw task completion, while structured UI yields cleaner outputs and better adherence to a prescribed persona. Building on this, they introduce Communication Policy Evolution (CPE), a prompt‑only loop that refines the policy during rollouts.\n\nThis matters because most deployments treat the agent’s output channel as an afterthought. By treating communication as a design variable, developers can extract more value from existing models without costly fine‑tuning. The hybrid and CPE methods point to a path for higher success rates using only prompt engineering.\n\nIn short, better chat formats may close the information gap that has long hampered autonomous LLM assistants.","[\"large-language-models\",\"human-computer-interaction\",\"ai-agents\"]","2026-06-15T04:00:00.000Z","2026-06-16T17:13:34.194Z","2026-06-16T17:13:37.002Z","published",null,[24],{"id":25,"reviewer":26,"round":27,"reason":28,"status":29},"editor-r1","editor",1,"Add a concise concluding paragraph summarizing the key finding and its implications for developers and users.","resolved",[31,32,33],"large-language-models","human-computer-interaction","ai-agents",[35],{"name":36,"url":37},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.14314",0]