Do Proactive Agents Really Need an LLM to Decide When to Wake and What to Anchor?
A new approach challenges the assumption that proactive agents must invoke LLMs on every user event. Rather than converting structured activity streams into text and asking language models to parse them back into decisions, researchers propose encoding raw event graphs directly with temporal graph learning models. This yields trigger probabilities and routing scores in a single forward pass, deferring LLM calls only when action is warranted. The shift from text-mediated reasoning to native graph processing reduces computational overhead while improving F1 scores across 14 model backbones, suggesting a broader architectural rethinking of how always-on systems should handle continuous signals.
Modelwire context
ExplainerThe deeper provocation here is not efficiency but scope: if trigger detection and context routing can be handled by a graph model trained end-to-end on raw event streams, then LLMs in always-on agents are being asked to do work that was never suited to them in the first place, not just work that is expensive.
This connects directly to two threads running through recent coverage. The story 'Do Language Models Track Entities Across State Changes?' found that LLMs defer and batch computation rather than updating state incrementally, which is precisely the wrong profile for a system that must respond to continuous event streams in real time. Separately, 'When Should Models Change Their Minds?' documented systematic failures in belief updating across extended interactions. Both papers were diagnosing symptoms; this paper proposes a structural fix by moving the always-on sensing layer out of the LLM entirely and reserving language model calls for moments where natural language reasoning is actually warranted.
The benchmark covers 14 model backbones, but the claim only holds if the graph-based trigger model generalizes to event distributions outside the training domain. Watch whether the authors or independent replicators publish results on a held-out activity schema, such as calendar or IoT sensor streams, within the next six months.
This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.
MentionsTemporal Graph Learning · Proactive Agents · LLM
Modelwire Editorial
This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.
Modelwire summarizes, we don’t republish. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.