Marty Cagan's stance, given in response to a request for a single AI tool that ingests all input sources (customer, sales, engineering, data) and auto-prioritizes the results: doing so would mean abdicating Product Sense to a tool, even though he acknowledges it is technically feasible with generative AI today.
The line he draws: generative AI can assist with synthesizing input from customers, sales, and engineering, but the judgment and prioritization function — deciding what matters and why — has to stay with the human product person. Related: AI-Fit Evaluation for Strategy.
Karina Stukan (CEO, Bizzy) draws a line between two uses of LLMs in stakeholder-facing product work. She uses a personal "internal agent" to aggregate and summarize recurring reports — monthly KPIs, CRM qualitative notes — which is a time-saving, low-risk use of AI. But she avoids using LLMs to simulate stakeholders (e.g., to predict how a given exec will react), on the grounds that understanding real people in the room can't be replaced by AI — it has to come from actually being in the room and reading them (see Presenting as Discovery).
Cagan expects generative AI to ease the dependency/Team Topology by Business Vertical (Durable Ownership Teams) problem and to genuinely help engineers and designers, but says he's most nervous about its effect on product managers specifically: AI can let a PM skip the thinking that coaching is meant to build in the first place, producing a plausible-looking artifact without the underlying judgment. See AI Removes 'Simmer' Time for Thinking.
Cagan reports reversing his own earlier advice to 'start with ChatGPT output and improve it,' after observing people submitting raw, unimproved AI output as their own thinking. His revised guidance: think through the problem yourself first, then use AI specifically to challenge and stress-test that thinking — not to originate it.
When using AI on a strategy or design problem, produce your own first draft before consulting the model, and use the model as a critic of your thinking rather than its author.
Julia Barham draws the same line for team-wide AI use, not just individual PM judgment: AI is useful for tasks like data synthesis, but it should be used with the team rather than instead of it, and should never substitute for the PM's own point of view or critical evaluation of AI output.
Barham's version of this caution, applied specifically to strategy synthesis: don't let an AI tool be the one that synthesizes customer, platform, and business data on the team's behalf. She schedules a recurring cadence — monthly or every two months — where the team synthesizes that data together and jointly discusses whether it changes strategy or existing hypotheses. The tool can assemble the data; the sense-making of what it means for the plan stays a team activity, not something to be delegated wholesale.