Beliefs

19 Tier-1 beliefs grouped by their parent theme. Each one has a sub-page with origin, evidence, and current state. The full graph (themes + Tier-2 / Tier-3 beliefs + projects + posts) lives in the knowledge graph.

19 beliefs · 11 parent themes · sourced from kg.json

Substance over hype

• Substance over hype

  Reduce any hyped category to its substrate before evaluating it. If the substrate is genuinely new, the excitement is earned. If not, the rebrand gets refused.

Agent-first

• Agent-first

  Build for autonomous agents as users, not just for humans. The serving lens is where the next layer of differentiation lives.

• Context over prompt

  Upstream context curation beats downstream prompt composition - the lever, not the I/O seam.

Breadth as differentiation

• Breadth as differentiation

  Cross-domain range is a genuine PM differentiator - the thesis that depth-axis refinement builds on and cannot replace.

• Breadth needs depth

  Post-ChatGPT, breadth without depth is Human-GPT - pick a niche and invest, axis is free, the choice is not.

PM taste

• Anti-customization

  Every customization setting in a PRD is a deferred design decision that will cost six months, an implementation partner, and day-zero go-live.

• PM is 99 Should-We Questions to Earn 1 Can-We

  Product management is a discrimination function: 99 refusals and prioritizations to earn the one rare can-we that actually moves the company forward.

AI PM skillset

• AI Fluency is Table Stakes for PMs, Not a Differentiator

  AI fluency is buy-in for application-layer PMs, not a flex - three load-bearing parts define the floor.

• LLMs are the primary daily tool in the stack

  LLMs top the daily-use stack across code, writing, data, learning. Factual observation, not extended-self mysticism.

Enterprise AI reality

• Enterprise AI production reality

  ~80% of enterprise AI experiments don't ship. The gap between a working demo and a working deployment is wider than vendors admit and more boring than analysts cover.

Second brain

• Self-instrumentation is how you debug your own thinking

  Instrument yourself the same way you'd instrument a system - name the patterns, externalize them, then synthesize them back.

• The second brain as context layer

  Plain markdown plus git plus a kg.json ontology - a bidirectional personal context layer that sits across every AI session and makes context-over-prompt operationally real.

Spec first, taste always

• Spec over Sprint

  When AI makes iteration nearly free, the binding constraint is spec quality - the model amplifies whatever you put in front of it, not how fast you put something in front of it.

• Taste over Execution

  AI commoditizes execution. Taste - knowing what to build and what to refuse - is the irreducible PM skill that no model can replace.

Career reflection

• Help the Market Flourish

  Grow the pie rather than fighting for share: the market-expanding posture compounds faster and costs less than conquest.

• IC path as legitimate endgame

  The Individual Contributor track is a destination, not a way-station - and AI augments IC reach more than management.

• Learn concepts, not tools

  Frameworks endure, tools rotate. Learn STP / JTBD / MVP / AARRR, not the SaaS-of-the-month.

LinkedIn as instrument

• LinkedIn as instrumental platform

  Posting closes the build-teach-learn loop - LinkedIn as instrument, not stage.

Personal projects

• Ship the Prototype

  The prototype is the position: a shipped, working version of an idea converts it from pitch to proof and closes the loop that makes personal projects compound.