hebel. today · 8 November 202637 / 73
Career · Product Management Evolution
Product teams debate the value of friction in AI-driven workflows
The case for friction. John Cutler argues that traditional product processes, such as multi-round document revisions and team discussions, contain positive friction that improves decision quality. He observes that teams often ignore these moments…
From 4 newsletters · 1 min read · one thing to try
The case for friction. John Cutler argues that traditional product processes, such as multi-round document revisions and team discussions, contain positive friction that improves decision quality (John Cutler from The Beautiful Mess). He observes that teams often ignore these moments of reflection as they turn toward autonomous AI workflows, potentially sacrificing long-term design viability for immediate deployment speed. Patrick Neeman supports this cautious approach by maintaining a strict separation between AI execution and human decision-making, insisting that tasks involving sensitive data or final signatures remain under human control to preserve trust (Patrick Neeman from AI for UX Substack).
The case for synthetic challenge. Agnes, a product lead in London, contends that AI-generated proposals often lack sufficient critical analysis because the tools are too agreeable (Techcubator). She implements a three-role system—writer, opponent, and judge—where separate AI instances simulate failure scenarios to stress-test product decisions before they reach a human reviewer. This method forces a rigorous evaluation of assumptions and trade-offs that might otherwise be overlooked in a fast-paced development cycle.
Synthesis of approaches. Both perspectives emphasize that AI-native product management requires new structures to replace the lost natural checks of manual workflows. While some teams prioritize preserving human interaction to catch errors, others build synthetic adversarial loops into their planning processes to achieve similar levels of scrutiny.
- PostHog now tags every analytics event to distinguish between human and agent activity, as agents currently generate half of the platform's dashboards (TLDR Founders).
- Product leaders at the 2026 Product Leader Awards prioritize building learning organizations that test and deliberately shelve AI projects lacking empirical evidence.
Try today · 45 min
Run a premortem on your next product feature using the three-role system.
Open three separate chat windows for the Writer, Opponent, and Judge roles.
You need
- A clear description of your feature, including what you are giving up and how you define failure.
Steps
- Draft your feature decision using the Writer role, ensuring you define the target outcome, the sacrifice, and the failure condition.
- Copy the decision into the Opponent role and prompt it to identify the top seven reasons the feature might fail within three months.
- Review the Opponent's output as the Judge, classifying each point as a change to the plan, a new validation task, or a risk to monitor.
- Update your final decision document based on the Judge's assessment.