Loading patterns…
Proactive Clarification & Active Disambiguation(PCAD)
Before planning or acting on an underspecified request, the agent first decides whether to ask at all (ambiguity detection), then generates the maximally informative clarifying question (framed as expected information gain over the plausible interpretations), and only then commits to a plan. Off-the-shelf LLMs recognize ambiguity when explicitly asked yet overwhelmingly default to a silent single-guess answer, so asking is an explicit trained or prompted behavior. Distinct from mixed-initiative-interface-patterns, which is about UI control-switching, and from human-in-the-loop, which approves an already-chosen action rather than disambiguating intent.
In 30 seconds
- What
- Agent detects ambiguity in a request, then asks one high-information question before acting, rather than silently guessing a single interpretation.
- When to use
- Requests are underspecified or vague, actions are costly or hard to reverse, and clarification can eliminate many wrong paths at once.
- Watch out
- Users may perceive repeated clarification as inefficient; balance asking against over-interrogation on genuinely obvious requests.
Ask the AI expert about this pattern
Opens the assistant with your question prefilled. You review it before sending.
Proactive Clarification & Active Disambiguation: Overview
Before planning or acting on an underspecified request, the agent first decides whether to ask at all (ambiguity detection), then generates the maximally informative clarifying question (framed as expected information gain over the plausible interpretations), and only then commits to a plan. Off-the-shelf LLMs recognize ambiguity when explicitly asked yet overwhelmingly default to a silent single-guess answer, so asking is an explicit trained or prompted behavior. Distinct from mixed-initiative-interface-patterns, which is about UI control-switching, and from human-in-the-loop, which approves an already-chosen action rather than disambiguating intent.
- Ambiguity detector decides whether a question is even needed
- Selects the question with the highest expected information gain
- One well-chosen question can split many interpretations at once
- Answers directly when the request is already unambiguous
- Prevents silent single-interpretation failures on vague inputs
- Explicit behavior since models recognize but rarely surface ambiguity
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.
References
The papers, specifications, and repositories this pattern is based on.
- Ask-before-Plan: Proactive Language Agents for Real-World Planning - Zhang et al. (ArXiv 2024)arXiv:2406.12639
- Modeling Future Conversation Turns to Teach LLMs to Ask Clarifying Questions - Zhang et al. (ICLR 2025)arXiv:2410.13788
- Knowing but Not Showing: LLMs Recognize Ambiguity but Rarely Ask Clarifying Questions - Su & Cardie (ArXiv 2026)arXiv:2605.25284
- InteractComp: Evaluating Search Agents With Ambiguous Queries (ArXiv 2025)arXiv:2510.24668
From the engineer behind this catalog
Get your agent architecture reviewed
This page documents one pattern. Your system runs dozens, and most failures live in how they fit together. Have the whole design reviewed against the 288 patterns in this catalog: architecture, reliability, evaluation and cost, every finding mapped to the pattern that fixes it.
€750 instead of €1,500, one week, written report and walkthrough call, until 30 September