Sherlock has always been able to answer on the fly. But "find our best leads and send them meeting requests" isn't a one-shot question, it's a project. Sending Sherlock off to handle multi-step or ambiguous requests without a check-in point meant crossing your fingers and hoping the AI understood the assignment. Not anymore.

Two updates land together to fix that: Plan mode, so you see Sherlock's approach before it moves, and multiple-choice questions, so Sherlock can ask for missing context instead of assuming it.

Plan Mode: See the Steps Before Sherlock Takes Them

When an organizer asks Sherlock something that needs deeper reasoning, or that involves Sherlock actually doing something on your behalf, Sherlock now stops to lay out its plan first.

How it works

1. You ask Sherlock a complex request (multi-step, action-oriented, or open to interpretation).
2. Sherlock detects it needs a plan and replies with a numbered breakdown of what it intends to do, formatted as a normal chat message.
3. You react to the plan:
   • Agree, and Sherlock proceeds to execute it.
   • Discuss changes, and Sherlock updates the plan and re-displays it for your confirmation.
   • Disagree, and Sherlock stands down. No action taken.

Key behaviors

• Nothing happens without your sign-off. Sherlock only starts executing after explicit agreement, no silent actions on your event.
• Plans are editable, not just approve/reject. Talk to Sherlock like you would a teammate: tell it what to change and it revises the plan in place.
• Same message format you already know. Plans show up as a regular Sherlock message in the thread, no new UI to learn.

Multiple-Choice Questions: Answer in a Tap, Not a Paragraph

The second piece: when Sherlock doesn't have enough context to give a solid answer, it now asks, with suggested answers you tap instead of typing out.

Key behaviors

• Single-select moves fast. Tap one answer and the flow jumps straight to the next question, or submits everything back to Sherlock if it's the last one.
• Multi-select waits for you. Tap as many answers as apply, then hit Submit to advance. The button stays disabled until at least one is selected.
• "Something else" always has a place. Every question includes a free-text fallback for the answer that isn't on the list.
• Ignore it and Sherlock adapts. Send a new message instead of answering, and Sherlock drops the question flow and answers your original request as-is, no dead-end conversations.
• Works everywhere Sherlock does. Same experience on the event mobile app and the web app.

Once every question is answered, Sherlock goes back into thinking mode and returns one refined answer that accounts for every response, instead of a generic best guess.

Built to Work Together

Plan mode and multiple-choice questions are designed to pair up. Ask Sherlock to identify and reach out to your best leads, and it might propose a plan (review the list, draft messages, send requests) and then tap you for the missing piece, like which seniority levels to prioritize, before it drafts a single message.

Why Organizers Will Love It

Exhibitor ROI and lead retrieval workflows live and die on trust: if you can't see what an AI agent is about to do, you don't hand it real work. Plan mode turns Sherlock from a black box into a collaborator you can steer mid-task. And multiple-choice questions mean neither your organizers nor your attendees have to write an essay just to get matchmaking, session, or networking recommendations that actually fit.

💡 Pro tip: Use plan mode for anything Sherlock will act on (sending messages, updating records), and let the clarifying questions do the work on anything ambiguous. Together they cut down on the "that's not what I meant" follow-up messages.