Convergence
Across PMM, growth, evals, and product, operators converge on the same shape: agents/automation own the inner loop (execution, retrieval, draft generation, experiment plumbing), and humans own the outer loop (alignment, taste, kill decisions, mission framing). The boundary is not "human-in-the-loop review" — it is a clean role split where humans never compete with the substrate on volume.
Operators
- Amol Avasare (Ramp) — Automate the four stages of a growth experiment; keep humans on alignment. CASH model: Collect, Automate, Scale, Hire — humans only stay where the AI cannot.
- Claire Vo (3x CPO) — Agents work when treated as a team, not a single super-tool and The unlock for AI agent productivity is management skill, not technical skill. Manage 9 agents like teammates; the unlock is management skill.
- Elena Verna — PLG sales-led companies need PLG, and PLG companies need sales — both, not one. PMM as substrate other functions self-serve from, not a gate they queue behind.
- Hamel Husain & Shreya Shankar — Evals are systematic data analysis on your LLM application — start with error analysis, not tests and Appoint one trusted-taste expert as the eval benevolent dictator — committees stall the loop. Evals are systematic data analysis with one trusted-taste judge — committees stall the loop.
- Jessica Fain (Pinterest) — Killing your own initiatives loudly is the highest-trust move with executives. Killing things loudly is the highest-trust upward move — taste is exercised, not delegated.
- Bret Taylor (Sierra) — Agents push SaaS from per-seat to outcomes-based pricing; the incentive flip changes everything. Agents push pricing to outcomes — the human pays for the alignment, not the per-seat license.
- Cat Wu — Taste is the scarce skill in an AI-native team. Taste is the bottleneck in an AI-native team.
- Boris Cherny — Underfund teams deliberately so AI substrate, not headcount, absorbs the work. Underfund teams so substrate, not headcount, absorbs the work.
Variation
- Avasare and Claire Vo describe the operating model (CASH, agent team).
- Verna and Boris describe the org-design move (substrate-as-PMM, deliberate underfunding).
- Hamel/Shreya describe the quality loop (evals = data analysis, single judge).
- Fain and Cat Wu describe the individual contribution shape (kill, name taste).
- The shared claim: the human's contribution is judgment + alignment, not throughput. The variation is which loop (growth, eval, launch, hiring) the operator scopes their boundary to.
Implication
For any AI-native function, name the inner loop (what the substrate runs) and the outer loop (what only humans decide). Resist the urge to put a human reviewer in every step — that re-introduces the queue. The PMM/PM/leader's job becomes alignment design (kill criteria, named taste calls, eval structure), not approval throughput.
Sources
- ins_cash-four-stage-growth-automation — Amol Avasare
- ins_agents-as-team-not-tools — Claire Vo
- ins_manager-skill-not-technical — Claire Vo
- ins_pmm-as-infrastructure-not-gate — Elena Verna
- ins_evals-are-data-analysis-on-llm-apps — Hamel Husain & Shreya Shankar
- ins_benevolent-dictator-not-committee — Hamel Husain & Shreya Shankar
- ins_kill-things-to-build-trust — Jessica Fain
- ins_outcomes-pricing-restructures-saas — Bret Taylor
- ins_taste-as-scarce-skill — Cat Wu
- ins_underfund-deliberately — Boris Cherny