We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Big Ideas
AI-agent UX is moving from disclosure to earned context. Scott Belsky’s first-mile framing treats progressive personalization—and staged requests for data—as the new progressive disclosure. Trust becomes an innovation variable: users may accept more privacy trade-offs when the return is clear, while agent-to-agent referrals may become a major onboarding path. He also puts personality, actionability, hospitality, and contextual memory alongside the graphical interface as core UX. The PM implication is to sequence permissions and personalization around demonstrated value, and to design the social handoff and memory experience rather than treating chat as the product.
Novel form factors need both a quality bar and a job hypothesis. Tony Fadell presents Apple’s foldable iPhone as a deliberate late entry: best-in-class hardware first, followed by developer-created experiences. Early user discussion points to two plausible jobs—media consumption and side-by-side productivity—while another PM frames the opportunity as closing the gap between “small internet” and “big internet” tasks, where habits and trust matter as much as technical capability. Treat those as hypotheses, not proof of product-market fit: instrument which users actually use the extra screen for after launch.
Tactical Playbook
Keep discovery human-led even when AI makes prototyping cheap. A practical small-team loop is: align with the organization on the time discovery deserves; interview ideal and non-ideal, paying and non-paying customers; synthesize recurring needs; revalidate them through non-leading conversations and surveys; then prototype with a deliberately mixed alpha group and iterate. Use AI to take notes, surface missed points, generate concepts, and accelerate prototypes—but not to replace customer conversations, because it lacks product-specific and customer-specific context. This matters because one PM-community report describes an AI research tool amplifying leadership’s existing beliefs while burying unusual feedback. Require direct customer evidence and an explicit dissent check before an AI-generated theme becomes a roadmap priority.
Case Studies & Lessons
Onboarding a complex product is context archaeology, not document retrieval. In a new role, one product practitioner built a “subway map” connecting roles, handoffs, artifacts, outputs, signals, and decision points, then converted it into reusable tables to pressure-test the end-to-end experience. Existing notes may be stale, weakly validated, or AI-summarized without real convergence; AI can point toward useful artifacts while still producing polished but incorrect understanding. Tickets were useful for delivery but poor for a newcomer; production code was the more definitive account of what the product actually does. For onboarding or a major product area, map one critical workflow, label live versus relic context, talk to an experienced operator, and verify the result against production behavior.
Career Corner
Replace mass applications with targeted proof. A reported job-search comparison produced five responses from 70 applications, versus eight replies and four interviews from 11 direct messages. Find two or three likely hiring managers through the company’s People page and recent team or product posts. Then send a short, company-specific work sample, a genuine point of overlap, or a low-friction coffee request. The message should ask an easy-to-answer question, demonstrate homework, and arrive before the application is buried.
Tools & Resources
Teresa Torres’ upcoming workshops focus on two increasingly important PM skills: AI Evals: The New Discovery Habit on September 23 and Story-Based Customer Interviews on September 24. The first offers an evaluation blueprint for personal workflows and customer-facing features; the second addresses how to choose what to build as delivery gets cheaper, using continuously collected customer stories.