CMU paper: planning agents should show plans, not ask preferences — up to 4.6x gains
rohanpaul_ai · x · 2026-11-17
A Carnegie Mellon paper formalizes ReSHAPE: planning agents shouldn't treat a user's initial request as fixed, but reshape the search space by showing concrete plans so users discover trade-offs they'd accept (e.g. a 3.7-rated restaurant to save money).
Analyzing 5,800 agent-revised specifications in travel and shopping with human-validated simulated users, the paper finds one correct trade-off makes the same planner reach an acceptable plan up to 4.6x as often; current agents admit at most 32% of acceptable plans vs over 52% with that trade-off alone; showing plans raises F1 by 9.7 points over asking preferences. Treat the request as a hypothesis to test against user reactions.
Related event: CMU paper: planning agents should show options, not ask preferences(2 posts)→
More from coding & agent
- Anthropic's daily code merges grew 12x in 4 months, now bottlenecked by human-scale infra — Dr_Singularity · 2026-11-17
- Kanban is a dead end for AI agents, argues dev pushing live reasoning-trace monitoring — kevinkern · 2026-11-17
- Software factory practice: verification infrastructure beats prompt and model choice — vikvang1 · 2026-11-17
- LuthorSQL: a 7B one-pass text-to-SQL gate that cuts agent API costs 33.3% — Ibrahim Memon · 2026-11-17
- Release Scout: A Third-Party Grok Bot Template That Reads Changelogs and Plans Your Bot Upgrades — SEO · 2026-11-17
- HTTP 200 doesn't mean success: production lessons from an Azure AI translation workflow — WirelessLife · 2026-11-17