---
title: "How can we develop transformative tools for thought?"
authors: Andy Matuschak, Michael Nielsen
year: 2019
venue: numinous.io (self-published essay)
kind: essay
status: note-written
captured: 2026-05-17
anchor_bucket: 6 · Individual essays & podcasts
---

# 2026-05-17 · Matuschak & Nielsen (2019) — "How can we develop transformative tools for thought?"

## What it is

A long-form essay from Andy Matuschak (independent researcher, formerly Khan Academy) and Michael Nielsen (physicist + writer) arguing that most contemporary "tools for thought" are not actually transformative. The essay both critiques the current state and proposes a research agenda for tools that change what humans can think, not just what humans can record.

Key concept: a **transformative tool for thought** changes the *medium* of thinking — not merely the *recording* of thinking. Spaced repetition is the example they keep returning to as a genuinely transformative pattern that almost no mainstream tool uses well.

## Why it matters for our line

This essay is the canonical contemporary statement of the "tools for thought" tradition. NOUS OS sits in this tradition — our L3 sub-line (personal knowledge co-evolution) is essentially asking *their* question for the AI era.

Even our L1 (Sandbox) is implicitly answering: "what would a transformative tool for *learning* look like?" Their answer is mostly individual-cognition-shaped (spaced repetition, executable explanations). Our answer is symbiosis-shaped (human + AI loop with explicit boundaries).

## Where we share

- The diagnosis that most current "tools" do not actually change cognition.
- The respect for representation-as-cognitive-substrate.
- The willingness to think in decade horizons.
- The use of detailed worked examples (their spaced repetition + executable books; our 20-minute Sandbox loop).
- The same anti-product, pro-research framing — they explicitly resist EdTech-style productization.

## Where we differ / what we add

| Matuschak & Nielsen | NOUS OS |
|---|---|
| Individual human + tool | Human + AI agent pair as the unit |
| Tools for thought (singular cognition) | Symbiosis with explicit boundary taxonomy |
| Spaced repetition as the canonical transformative pattern | Co-evolution loop as the candidate transformative pattern |
| Empirical evidence is sparse and aspirational | Insistence on per-session evidence + N-explicit reporting |
| Largely silent on AI / LLM era | Centers LLM-era misalignment risk in boundary design |

The most important difference: their essay predates the LLM era's *confidence* problem. Their tools enhance individual cognition where the tool is dumb (cards, simulations, executable papers). Our tools must enhance individual cognition where the tool is articulate, persuasive, and often confidently wrong. That changes the design.

## What this changes in our practice

- **L3 methodology should borrow heavily from their decade-horizon framing.** A 90-day Obsidian segment + retrospective task is too short for Matuschak-Nielsen's standard; we should either acknowledge this honestly or plan multi-year cohort designs.
- **Their detailed-worked-example style is a model for how to write up Sandbox sessions.** Our review packets should look more like *Quantum Country* (their executable explanation work) and less like clinical trial reports.
- **Their willingness to publish before evidence is in hand** is permission to do the same — provided we are honest about what is conjecture vs measured.

## Limitations of this work (from our perspective)

- No empirical study; the essay is conceptual + survey.
- Pre-LLM (2019); the AI confidence problem barely figures.
- Heavy on individual cognition; almost silent on pair / relationship cognition.
- "Transformative" is undertheorized — when *is* a tool transformative vs merely useful? They gesture but do not measure.

## Open questions for follow-up

- Has Matuschak published a post-LLM update to this framework? (His patreon may have it.)
- Does Nielsen's *Reinventing Discovery* have material we should add?
- What is the relationship between "transformative tools for thought" and our "compounding wisdom" north star? Are they the same thing under different vocabulary?

## Citation

Matuschak, A., & Nielsen, M. (2019). How can we develop transformative tools for thought? *numinous.io*. https://numinous.io/papers/tools-for-thought.html