← Back to Discover

Unreviewed draft. Not published content.

DISCOVER / AI × PKM

Progressive Summarization in the Age of AI Summaries

Date
Type
Method note
Source
Tiago Forte — Progressive Summarization (Forte Labs)
Tags
AI × PKMAI Second BrainTools for Thought

AI can summarize almost anything in seconds. That is useful—and it is also how people accidentally build a graveyard of fluent, forgettable notes.

Tiago Forte’s Progressive Summarization was designed for a different problem: not “can I get a short version now?” but “will Future Me find and trust this when it actually matters?” That design still holds when models write the first draft of a summary.

This Discover piece explains the layer method, how it sits next to PARA and CODE, and how to use AI as a Distill assistant without outsourcing judgment. It is the human-side counterpart to Discover #1 (LLM Wiki pattern) (machine-maintained wiki): here the goal is notes you make discoverable.

Verification status: Not tested by Cyber-Taoist editorial. Claims follow the cited sources. Live AI demos (including Forte’s) are not re-run here.


The real job: knowledge through time

Author view (Forte): Capture is easy. Moving content between apps is easy. What is hard is moving knowledge through time—so that a book, a call note, or a research rabbit hole shows up when a future project needs it, not when you first felt inspired.

A second brain is an external store that packages knowledge into packets Future You can review, use, or delete. Memorization techniques and endless “infovore” consumption miss the point if the timing is wrong: you often read the right idea at the wrong moment.

Editorial analysis: AI makes that timing problem worse if you treat every chat summary as finished knowledge. Fluent compression without a retrieval design just accelerates note cemeteries. Progressive Summarization is a product design for Future You, not a party trick for Past You.


The design tension: discoverability vs understanding

Author view (Forte): Every note balances two priorities:

  • Discoverability — small, simple, digestible → needs compression
  • Understanding — details, examples, sources → needs context

Compress too hard and the note means nothing a year later. Keep every detail and Future You will not risk the time to re-read it. Progressive Summarization is opportunistic compression: add layers in small spurts, usually when you are already reviewing the note for live work—not in a separate “organize Sunday.”

The layers

Layer What it is
L0 Full original source
L1 What you first bring into the note (quotes, clips, rough capture)
L2 Bold the best parts of L1
L3 Highlight the best of the bold
L4 Your own informal executive summary at the top—in your words
Remix Rare: turn the idea into a post, sketch, deck, or other output

Earlier layers stay intact. That safety net lets you summarize boldly without fearing irreversible loss.

Editorial analysis: L4 is the line AI should not fully own. A model can propose candidates for L2/L3; you should still write (or heavily rewrite) L4 if the note must carry your judgment. That is the difference between Distill assistance and outsourced thinking.


PARA: where it lives; PS: how it is shaped

Author view (Forte): PARA—Projects, Areas, Resources, Archives—sorts notes by actionability, not by academic topic. Placing a note in a Project schedules a short review horizon; Areas less often; Resources when you choose that topic; Archives cold storage.

Progressive Summarization does not replace PARA. PARA answers “where does this packet sit?” PS answers “how is the packet designed so Future You buys the review?”

Editorial analysis: Four empty folders are not a second brain. If you only Organize and never Distill or Express, PARA becomes interior decoration. Pair folder placement with at least one intentional layer pass when a note earns a revisit.


CODE in an AI stack

Author view (BASB / Forte): The workflow is Capture → Organize → Distill → Express. Common failure mode: endless Capture and Organize with no Distill or Express. AI can speed Distill and help maintain prompts; Express—shipping a claim, a decision, a creative piece—stays human.

Editorial analysis — practical split with Discover #1:

Need Lean on
Multi-source research that must stay consistent over months LLM Wiki pattern (#1): model maintains the middle layer
Personal notes that must sell themselves to Future You Progressive Summarization (#2): you deepen layers
Fast first pass on a long article AI draft of L2 candidates → you bold/highlight/own L4

Use AI to propose compression. Use PS discipline to accept, reject, and voice it. If every summary is model-authored and never revisited, you have RAG-with-sentiment, not a second brain.


A one-week practice (no new app required)

  1. Pick five notes you already reopen for real work (not random archive).
  2. On each revisit, add only one next layer (L2 or L3)—do not rebuild the vault.
  3. For the one note that actually unlocks a decision, write L4 in your own words (even if AI drafted a summary first—rewrite it).
  4. Optional: ask a model for “candidate bold sentences,” then delete half before you bold.
  5. End the week by Expressing once: a short claim, checklist, or message that uses one L4.

Editorial analysis: Opportunistic beats ceremonial. A weekly “summarize everything with GPT” ritual fights the method.


Limits and failure modes

From sources / editorial:

  • Classic Progressive Summarization examples skew Evernote-era; the layer logic transfers to Obsidian, Notion, Apple Notes—the UI does not.
  • BASB is also a course/product ecosystem; evaluate the method, not the funnel.
  • AI Second Brain demos are illustrations; this article did not re-test them (verification: none).
  • Over-automation risk: perfect L2/L3 with no L4 → discoverable sludge without ownership.
  • Under-automation risk: refusing all AI help and drowning in L1 capture.

Bottom line

AI summaries optimize for now. Progressive Summarization optimizes for later—for a skeptical Future You with seconds to decide whether a note is worth opening. Keep the layers. Let models suggest compression. Write the executive meaning yourself. That is how Distill stays a skill instead of becoming a subscription.

Together with an LLM-maintained wiki (#1) and a local discovery layer (#3), you get a minimal triangle: compile, distill, find—without pretending any corner replaces judgment.


Sources

  1. Tiago Forte — Progressive Summarization (Forte Labs)
  2. Tiago Forte — The PARA Method
  3. Building a Second Brain · BASB overview
  4. Background only: Second Brain Playbook 2026 (second-hand; not primary)
← Back to DiscoverBack to top ↑