Every company on your list says AI. You need to know which ones have operationalized accountability before you spend positioning energy. These three questions work in any first conversation.
"When something the AI does turns out to be wrong — where does that go?" Listen for an incident path. Companies that have operationalized AI name a team, a process, a change that happens before the next run. Vague answers about model improvement mean nobody owns failure.
"Walk me through a workflow where AI takes action — at what point can a human stop it before it's committed?" Listen for whether they distinguish reversible from irreversible actions. Human gates belong at the commitment threshold, not after.
"Who decided how AI gets used in core workflows — was design part of that decision, or did design come in after?" The mandate question. You're learning whether this role shapes the operating model or decorates one that Legal, Product, and Engineering already locked.
Ask all three. Concrete answers on all three mean they've thought about AI as something to govern. Abstract answers on all three mean they're hiring you to make AI features look trustworthy.
Reading the answers
Q1 — Failure routing
- Strong: They name a role or process that owns AI errors. They distinguish model limitations from operating failures.
- Weak: "The model is getting better." No named owner. Errors treated as training data, not incidents.
Q2 — Human gates
- Strong: They know which actions are reversible vs. irreversible. Checkpoints exist at commitment thresholds, not confirmation modals.
- Weak: "Users can always undo." No reversibility gradient. A confirmation step framed as UX choice, not operating requirement.
Q3 — Design authority
- Strong: Design was in the room for operating-model decisions. They can name who owns AI design standards and who can block a launch.
- Weak: "Design and product collaborate closely." Adjacent to governance, owns none of it.
Calibration: Ramp and Headway already post with accountability language in their listings. Use those calls to validate the pattern. Companies posting only tool-fluency language ("familiarity with LLMs, generative design tools") are telling you where they are.