We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Big Ideas
AI’s next surface is ambient micro-work. Hiten Shah argues that autocomplete, extraction, classification, cleanup, and routing are often “over-modeled”: small local models are already good enough for many of these jobs, especially when they need to be fast, private, and always available. He describes the next product surface as work too small to ask for—finishing a few words, fixing a sentence, extracting a fact, or classifying something as it arrives—and says local AI should disappear into the software rather than expose model setup to users. Scott Belsky’s shorthand is that agent personality, proficiency, and personalization are becoming the new UX.
AI is expanding PM self-service, but context and validation remain the boundary. One PM reports getting usage metrics in about five minutes instead of sometimes waiting a week, and answers about features, code, databases, analytics, and logs in five to 10 minutes instead of several weeks. That speed depended on analytics coverage plus read-only access to the codebase and a database copy; the agent also surfaced existing but hidden product functionality. The practical limit is clear: machine-readable context works well for analysis, while agent-generated changes are riskier when the codebase is stale or a change touches shared state or core flows. A product-craft proposal offers a useful division of labor: keep strategy, hypotheses, metrics, success validation, and trust-building human; automate synthesis, monitoring, and research. Messy data, regulated environments, research context, and model accuracy still demand expert review.
Tactical Playbook
Turn “grow” into a choice architecture. For a PM inheriting a neglected portfolio, use Roger Martin’s Playing to Win sequence: define point A (current revenue or users) and point B (the growth target); identify the gaps; choose non-contradictory actions to close them; then test those choices with internal and external stakeholders. Add a field-level scan of the customer journey and after-sales experience, stakeholder alignment, competitors, market size, and company portfolio before budgeting. Look for a distinction that makes a target segment prefer you, but check whether that niche can reach the company’s ambition; iterate until the choices and expectations align.
Move roadmap tradeoffs upstream. A roadmap that lists commitments without showing what each one delays or makes harder makes every item look independently reasonable. Capture those opportunity costs and the reasoning in planning, BRD, or PRD material; keep the shared roadmap as a clean statement of direction rather than asking it to carry the full decision debate.
Case Studies & Lessons
A feature gap became a design partnership. At Rippling, a customer wanted Excel- or Google Sheets-style financial calculations that the platform did not yet support. The technical product manager acknowledged the gap, escalated it to product with a concrete date for an answer, offered an interim workaround, and kept the customer involved. The customer ultimately helped design and architect the feature, becoming a design partner rather than an adversarial escalation. The repeatable play is: separate what works from the one product gap, avoid promises without product confirmation, give a dated next update, and keep communicating even while the answer is pending.
Career Corner
AI PM interviews reward problem selection over AI vocabulary. Advice from a current AI PM is to distinguish problems that genuinely need AI from processes or systems that simply need improvement. A candidate who recently received two AI PM offers says interviewers focused on whether she could identify a real user problem, define AI’s limits, phase the work, understand the workflow, and compare the bet with the existing roadmap—not on advanced model theory. Prepare two cases in that format, while learning where MCP, RAG, APIs, and the company’s cloud architecture matter.
