We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Big Ideas
Ravi Mehta’s revised framework names 12 competencies a product team needs; his argument is that AI changes how the work is done more than the role’s overall shape. No single person is great at all 12. For example, product definition can start with working prototypes rather than static specs; delivery is shaped as the product emerges with humans and agents; and quality becomes ongoing as models personalize and drift.
Treat this as a team-design problem, not a mandate for every PM to become “full stack”: individuals should have distinct strengths while the team covers gaps. With build capacity abundant, outcome ownership means directing it at the right business results; strategic impact compounds those results rather than accumulating features.
Tactical Playbook
Design agents for workflows, not just chat. In an a16z discussion, speakers argue that when software needs a discrete choice, a model can interpret free-form text and select from predefined options—often faster, cheaper, and more accurately than generating prose. They also caution that chat is not automatically an efficient interface. For agents that take action, they call for tracking authentication and API activity and scoping permissions by resource and operation, such as read/write access to one folder and read-only access to another.
Add controls and exception paths. A PM-automation checklist distinguishes the trigger, worker, and instructions from three often-missing pieces: an independent output grader, a gate before proceeding, and state that remembers what happened. Its advice: audit any live automation that has not been checked since it was built. A concrete PM use is a weekly Salesforce scan that surfaces deal opportunities for PM support through explanation—not a new feature—and roadmap learnings. Hiten Shah’s related warning is that companies want agents but often have not documented how their best people decide; process covers the normal case, while experts know what to do when reality diverges. Capture that judgment and define which exceptions need human input.
For launches, one founder recommends a narrow audience, one concrete action, and a short feedback loop: check next-day completions rather than visits, then revise landing-page or onboarding flows within a day. Their anecdote: repeated conversations later brought 1,000+ organic users despite some posts staying below 1,000 impressions.
Case Studies & Lessons
A YC robot-agent discussion illustrates where to use model flexibility: repeated, proven actions can run as code or reusable skills, while a vision-language model handles variable steps such as object detection and failure recovery. The speakers identify latency in an always-on model loop as a constraint that can make the approach economically impractical. For product teams, reserve adaptive reasoning for uncertain branches; do not pay its latency on every routine step.
Career Corner
Mehta’s self-assessment makes development concrete: mark up to three competencies as strengths and three for focus, then compare notes with a manager before the next one-on-one.
For a senior PM taking on direct reports, a practitioner discussion recommends an organization-dependent player-coach model: take a frontier or high-risk area as a pilot, then give PMs end-to-end ownership of more familiar work and room to shape strategy in their areas. The trade-off is fewer projects for the lead; a former report warns that prioritizing IC work can crowd out coaching and communication.