When designers, engineers, and product managers prototype AI features together, who ends up owning the prompt?

This explores how the traditional handoff between designers, engineers and PMs changes when they prototype generative AI features side by side. The clearest answer in the collection is that the prompt becomes a shared workspace, and that makes role boundaries blur. In a study of 39 practitioners across 13 design sessions, prompts kept breaking into the same four parts: input context, system instructions, output constraints and few-shot examples. Different roles gravitated to different parts, so nobody owned the whole thing. Engineering didn't simply receive a finished design How do teams decompose prompts when collaborating on generative AI?. The handoff turns into co-authorship of one shared artifact.

The second shift is that designers start doing work that used to need an engineer. Canvil, a Figma widget for writing and testing system prompts, shows designers shaping model behavior directly by tinkering with it. That puts their judgment about users inside the model adaptation itself, where it used to arrive as requirements passed downstream Can designers shape LLM behavior without deep technical knowledge?. Dubberly gives the historical reading: when LLMs become a design material, design for agents starts to absorb interaction design. That is the same pattern as earlier technology shifts that redrew discipline boundaries Is design for agents becoming the next major design paradigm?.

If designers can tinker with models, does design expertise still matter? Lam and colleagues argue it matters more when the system's design decisions stay visible. An architecture that shows how the AI generates and evaluates output at each interaction builds ongoing design judgment into the product, so HCI skill can't be engineered away Does visible design structure keep HCI expertise necessary as AI advances?. Kat Holmes at Salesforce names the new skills the merged roles need: experience architecture, agent collaboration, conversation design, and context maps and ontologies What design skills do teams need for the AI era?. Several of these sit between traditional design and engineering, which may be why the boundary is moving.

One lateral warning comes from research on teams of AI agents. Teams with diverse viewpoints beat solo ideation only when each member has real domain expertise. Diversity without expertise did worse than a single competent agent Does cognitive diversity alone improve multi-agent ideation quality?. That study is about AI agents, not people, so treat it as a hint rather than evidence about human teams. The takeaway for cross-functional prototyping: blurring roles helps when each person still brings deep knowledge from their own discipline, and hurts when it turns everyone into a generalist.

The collection doesn't say much about the PM's role in particular. None of these notes shows how product management changes when prompts become the shared prototype, and that's a real gap. The evidence that exists points one way: the prompt, not the mockup, is becoming the place where the roles meet.