Speak for Me: Giving LLMs the Situational Awareness to Participate in a Meeting
Muneeb Khan, Frederic Kirstein, Terry Ruas and Bela Gipp find that prompt-only meeting delegates stay silent on 51.4% of the moments they should have spoken, and build CAPA, an architecture whose separate modules track state, forecast, decide and phrase.
Ask this paper
Knowing when to speak is the actual task: with no structured way to track stances, coverage and the floor, a delegate misses half the opportunities regardless of how well it writes.
Six named components: a Perceiver that updates meeting state per turn, a Predictor that forecasts continuation, a Controller that decides whether to speak and which proposition to surface, a Generator that phrases it in the participant's style, two judges, and a Recalibrator that folds verdicts back into state.
Silence drops from 51.4% to 2.5%, credited recovery doubles from 26.1 to 52.2, and hallucination stays at 0.6% across 137 AMI meetings.
Meeting state is the lever: ablations show the explicit state is what closes the recognition gap, and raw-context scaling alone does not, which is the transferable finding.
The failure mode moves rather than vanishing: it shifts from omission to selection, with each residual near-miss attributable to a specific module, which is what a well-factored architecture buys you.
Abstract
In online meeting delegation, LLM agents fail to recognize when to speak. With no structured way to track stances, coverage, and floor, they miss the moments where they should contribute. Prompt-only delegates stay silent on 51.4% of the absent participant's talking opportunities on the AMI corpus. We present CAPA (Collaborative Agent Predictive Architecture), an architecture for online meeting delegation. A Perceiver updates the meeting state from each observed turn. A Predictor forecasts how the conversation will continue. A Controller decides whether to speak and which proposition to surface. A Generator phrases the chosen contribution in the participant's style. Two judges score the forecast and the action against the next observed turn. A Recalibrator updates the meeting state from those verdicts for future decisions. To evaluate online delegation, we introduce an episode-level protocol that scores whether, when, and what a delegate contributes around the participant's actual idea units. The protocol's schema-constrained LLM judges align with human annotations at Cohen's kappa = 0.71. On 137 AMI meetings, CAPA reduces the silence rate from 51.4% to 2.5%, doubles credited recovery (26.1 --> 52.2), and keeps hallucination at 0.6%. The failure mode shifts from omission to selection, with each residual near-miss attributable to a specific module of the architecture. Mechanism ablations identify the meeting state as the lever that closes the recognition gap, where raw-context scaling alone does not.