PKM Wiki

Source - YB 2026 - Context Compounds

2 min read

Citation

YB, "Context Compounds," Engineering Agency (Substack), February 16, 2026. URL: https://engineeringagency.substack.com/p/context-compounds

Overview

Argues that as AI agents become commodity infrastructure, the scarce competitive resource is personal context — rich, accumulated, structured documentation of the user's life, thinking, and decision-making patterns. This personal context "compounds" over time in the sense that marginal context added on top of existing context produces disproportionately large gains in agent effectiveness, because agents can now reason across accumulated material rather than just about the latest addition.

Main Arguments

  1. Context as the new moat. Public data and model quality are commoditizing; personal context is not. "Personal success depends on feeding [agents] rich contextual data about your life."
  2. Quality over quantity. Agents work when they understand cognitive patterns — "how you connect ideas and how you process new information" — not when they have raw data volume.
  3. Trust via alignment. Personal agents can only be trusted to act on your behalf when they have enough context to "think and act like you." Without this, delegation is unsafe.
  4. Accessibility. Context can be private; local markdown files suffice. You do not need a public footprint to build useful agent context.
  5. Example: Tyler Cowen actively documents missing biographical material so that future AI systems have better data about him — an explicit context-building practice by a prominent public intellectual.

The Core Concept

Context compounds in the specific sense that:

  • Each added note about your life/thinking is useful on its own (linear)
  • But combined with existing notes, new notes enable cross-reference, trend-detection, and pattern-matching the isolated note could not produce (super-linear)
  • Over time, the accumulated corpus becomes a progressively richer substrate — agents operating on it become progressively more effective

Practical Recommendations

  • Voice notes: daily, under 5 minutes, captured in voice mode
  • Evening reflection: 10-minute journaling or AI-guided daily call
  • Process documentation: document entire project workflows end-to-end
  • Philosophical tip: keep it simple; elaborate systems become burdensome and stall

Why This Matters for PKM

This piece is the explicit practitioner articulation of a shift in what the vault is for:

  • Classical PKM: vault as external memory for the human user
  • Contemporary PKM (YB's framing): vault as substrate that AI agents read from, enabling them to act on the user's behalf

The compounding claim positions consistent daily capture as a high-leverage investment because the marginal value of each new entry depends on the corpus it lands in. This converges with broader compounding knowledge arguments but adds an AI-substrate twist.

Concepts Introduced or Amplified

  • Context compounds (the central claim)
  • Context as moat (competitive resource shift)
  • Agent-readable vault (vault framed as AI substrate)
  • Private-context-first documentation (useful without publication)