Four decisions carry the screen. The agent shows its plan before its answer, with a count next to each source — 142 profiles narrowed to 38 to 24 — so the user can see how the number was reached instead of trusting a black box. Results land as a structured table inside the chat, with company, funding, email, and a match score in scannable columns, and only the top 3 of 24 shown so the conversation stays readable while the full list stays one click away. Suggested next actions sit directly under the result, turning the usual dead end after an answer into the actual workflow: generate a cold email, push to HubSpot, find lookalikes. And connected data sources are pinned to the composer, so the user can see and change what's being searched at the moment they're asking.