We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Big Ideas
AI is accelerating product execution—not product judgment. Linear’s first data report, based on 127,000+ paid users, found that AI adoption more than doubled across every function from January to June: PM usage rose from 12% to 34%, AI authored nearly half of all issues, and pull requests increased 111% over two years. Teams using coding agents went from 21 to 65 pull requests per week, versus 8 to 10 for teams without them. Yet time spent on customer requests, documentation, and projects stayed flat across functions; AI work appeared as a new layer, and total work increased rather than decreased. Treat discovery, prioritization, and judgment as the bottleneck to protect—not capacity that automatically disappears when execution gets faster.
Agent-first UX will require permission rails, not just better browser automation. Scott Belsky’s “Favored Agent” thesis is that services will eventually give trusted agents direct, preferred access. His real-world example shows the gap: Instinct’s agent got stuck when ShopPay required confirmation, could not ask for verification, and stored a CVV in a way that conflicted with card-network rules; Belsky suggests agentic payment tokens or a direct ShopPay connection instead. For consumer and commerce products, design agents as authenticated actors with explicit permissions, human handoffs, and service integrations—not as humans navigating a CAPTCHA-filled interface.
Tactical Playbook
Use multi-agent graphs when the failure mode is a wrong fact. Aakash Gupta compared four graph designs with single-shot prompts across 40 PM tasks; the strongest applications included PRDs, pricing analysis, metric investigation, roadmap prioritization, opportunity sizing, and support-ticket synthesis. The winners did not create better taste: they recounted tickets, re-derived calculations, attacked plans with premortems, applied finance gates, and forced disagreements back to evidence. Graphs lost on positioning, naming, and strategy bets, where a second agent became an expensive yes-man. The operating rule is simple: fact risk → add independent checks; opinion risk → use one pass and apply human judgment. Budget accordingly: graphs cost roughly 4–6× the tokens and 5–15 minutes versus 1–3 minutes for a single pass.
Case Studies & Lessons
Daunt’s bookstore turnaround puts data in its place. In discussing the Waterstones and Barnes & Noble turnarounds, James Daunt says data-led predecessors overallocated space to fast-selling board books and underallocated it to young readers, leaving both groups frustrated. He argues that centrally held data can impose uniformity, while local stores should first learn how they engage their own customers and only then reintroduce data as a useful reference. His operating model reinforces that choice: flatten hierarchies, create collective responsibility, and acknowledge mistakes quickly without assigning blame. For product teams, give customer-facing teams real decision authority and use metrics to improve local judgment—not to erase it.
Career Corner
Design-to-PM transitions need evidence of product ownership. A director who moved from design into product said external interviewers still saw “a designer with an inflated title.” The practical repositioning advice is to make the resume roughly 90% outcomes and metrics, emphasize roadmap, prioritization, and trade-offs in interviews, and ask an engineering lead or CEO to verify that you owned product calls rather than only execution. The commenter also warned that even after these changes, about half of interviewers still saw a designer—so this is a positioning tactic, not a guaranteed fix.
Tools & Resources
Lighthouse makes delivery uncertainty legible. The free, open-source, self-hosted tool imports history from Jira, Azure DevOps, Linear, ServiceNow, or CSV and uses Monte Carlo simulations to produce 85% and 95% completion dates. It also exposes scope growth, percentile trends, and blocked work instead of reducing delivery to a single negotiated date. The community tier supports up to three teams and one portfolio; self-service and enterprise tiers use the same product. It is worth testing when stakeholders need a forecast they can interrogate rather than a promise disguised as precision.