Pitching an LLM-for-Autonomy Initiative to Leadership to Win Scope

After a long discussion about applying large language models to autonomous driving, the interviewer asks a pointed follow-up: **suppose we now want to win scope for this idea. How would you present it to leadership?** The idea is to bring LLM or vision-language capabilities into a self-driving program. Leadership in a safety-critical organization will be skeptical because of hallucination and real-time concerns, and other teams may already own adjacent areas such as data tooling, simulation, or planning. Explain how you would frame, support, and pitch the proposal so that it earns resources and ownership, and draw on a comparable time you won support for an initiative if you have one. ```hint Start from their problem What does leadership already worry about that this idea would measurably improve? ``` ```hint Shrink the first ask What is the smallest, lowest-risk version you could prove quickly? ``` ### Clarifying Questions - Who is the audience: your direct manager, a director, or a cross-functional review that includes the safety organization? - Is the scope being sought new headcount, ownership of a new area, or permission to run a pilot? - Does another team already own the area the idea touches, such as data tooling, simulation, or planning? ### What a Strong Answer Covers - Framing around a problem leadership already prioritizes, with a measurable target - Evidence such as a prototype or a data result, not only an argument - A phased, low-risk first step with explicit success and kill criteria - Addressing safety and hallucination concerns head-on, including what the system will not do - Pre-alignment with stakeholders and handling of overlap with teams that own adjacent scope - A concrete ask (people, compute, time, decision date) and how progress will be reported ### Follow-up Questions - Leadership says no. What do you do next? - The planning team sees your proposal as encroaching on its area. How do you resolve that? - Your pilot hits its quality target but increases compute cost. How do you present that trade-off? - What result would make you recommend shutting the initiative down?