Josh Elman argued on September 14, 2026, that AI has shifted product management from writing specifications before development to building and testing prototypes early. He said the lower cost of making demos increases the importance of deciding what belongs in a product, because working products still require substantially more effort than prototypes.
Elman, who began as an engineer at RealNetworks and later joined LinkedIn after interviewing with Reid Hoffman, described how his view of a product manager’s central artifact changed from the specification to the story. At LinkedIn, he wrote a 120-page specification for a jobs platform within the social network, but later concluded that a system description was less important than a clear, repeatable account of who would use the product and why it would matter in their lives.
In Elman’s account, the traditional development loop moved from idea, specification, costing and scoping to building. AI reverses much of that order: teams can first build an idea quickly, play with it, then determine the visual, user-experience and engineering design needed to turn it into a real product before shipping and learning. He cautioned that cheap demonstrations should not be confused with production-ready systems and framed product selection as an impact debate between alternatives rather than a resourcing debate between building something or nothing.
Elman said the durable work of product management is defining a product’s purpose, its core user actions and the expected cycle for those actions. He warned against treating figures such as daily-to-monthly active-user ratios, signups, waitlists, annual recurring revenue, token volume or App Store rankings as sufficient proof that people are really using a product. Instead, he recommended examining direct traffic and whether users perform the core actions. For AI products, he said prompt and conversation transcripts can expose user expectations, failed interactions and moments when users rephrase or abandon a task, but product managers should read those records rather than delegate all interpretation to AI summaries.
Onboarding remains central to that user story, according to Elman. He recommended designing for curious users between highly motivated adopters and casual visitors, using simple, discrete steps to explain the product’s purpose and capabilities. He characterized a blank AI prompt box as a weak onboarding experience when it offers no concrete guidance, and advised helping users reach at least one valuable use case, ideally with their own data. He also said onboarding tests should be judged by later retention and core actions, not merely completion rates.
Elman cited Twitter’s onboarding redesign after he joined the company in late 2009. Although millions signed up amid heavy public attention, many did not return because the service’s purpose was unclear. Twitter’s Learn Flow instead introduced tweets, following and the timeline one concept at a time; Elman said it improved retention more than anything else the company shipped that year. His conclusion was that AI should accelerate prototyping without accelerating judgment: product managers must still decide what the product does for users, keep the experience coherent and determine whether people are genuinely using it.