The Structure Has Changed
For a decade, product and growth teams optimized the same funnel: surface the feature, nudge the user, measure activation. Tooltips, onboarding checklists, in-app modals — the entire discipline of feature discovery was built on one assumption: users need to find the capability.
That assumption is now structurally broken.
When an AI agent mediates the product, users don't navigate to features — they state a goal and the agent picks the function. As Elena Verna argues, the moment a user delegates to an agent, the old discovery layer disappears entirely. No tooltip survives that handoff.
For operators at non-tech companies who are right now deciding how to deploy AI inside their GTM and product motions, this isn't a future-state concern. It's this quarter's adoption problem.
The New Bottleneck Is Trust, Not Friction
If users won't delegate to the agent, the feature never gets used — regardless of how well it's built or how clever the onboarding flow is. The constraint has moved upstream.
This reframes what growth and RevOps teams should actually be measuring. Activation rates, feature adoption breadth, tooltip click-through — these metrics are answering the wrong question. The question is now: does the user trust the agent enough to hand off the task?
Trust here is specific and operational. It means the agent produces outputs the user doesn't feel compelled to double-check. It means prior handoffs didn't embarrass the user in front of a customer or a manager. It means the agent's scope is legible — users understand what it will and won't do.
Building that trust is not a UX problem. It's a reliability, transparency, and feedback-loop problem. Teams that treat it as a copy-and-tooltip exercise will watch adoption plateau and blame the AI.
The Trap Most Companies Are Already Walking Into
The critical failure mode Verna identifies is companies treating AI as a feature — adding a ✨ icon to an existing workflow and calling it transformation. That approach doesn't eliminate the discovery problem. It creates a new one: now users have to discover the AI wrapper on top of discovering the underlying capability.
The product has to be the AI, not contain it. For operators, that distinction has immediate implications for roadmap prioritization, success metrics, and how you structure rollout.
Actionable takeaway: Audit your current AI deployment against one question — does the user still have to find and invoke the capability manually, or does the agent surface and execute it based on stated intent? If it's the former, you've built a feature, not an agent experience. Redefine your adoption KPIs around delegation rate and task completion confidence, not clicks. That's where the leverage is now.