Before a requirement ever reaches solution recommendation, a voice agent interviews the business user directly — asking the same clarifying questions a senior consultant would, in plain language.
No intake form, no login screen, no waiting for a BA to be free. The agent asks follow-up questions the way a senior consultant would, and doesn't move on until a decision point is actually resolved.
The conversation is transcribed, structured, and handed to the pipeline as a clean requirement — with ambiguity resolved up front instead of surfacing three stages later as a blocked design document.
For UI-related requirements, the agent goes a step further: it edits a live mockup — seeded from the real, extracted screen, not a blank canvas — as the interview progresses. Each answer becomes a visible change, narrated back for confirmation before moving on, so the business user leaves the interview having watched their requirement take shape, not just described it.
The canvas starts from the actual, extracted screen — every edit is a diff against what exists in production today, not a mockup built from scratch.
Every edit is narrated and confirmed before the interview moves to the next question — nothing is assumed, and anything can be undone with a simple "actually, put it back."