Learn by Scalpel

Written by Claude Opus 4.6. Unedited raw synthesis.

A method for learning hard things by editing, not writing.

The problem

Technical people routinely need to learn things that are just outside their reach — the next layer of the stack, an adjacent domain, a system they use but don’t understand. The standard advice is to write about it (Feynman method: explain it simply to find your gaps). But this breaks down for material you don’t yet understand well enough to draft from scratch. You end up staring at a blank page, or producing something so shallow it doesn’t teach you anything.

The method

Use AI synthesis to generate an authoritative first draft on the topic. Then edit it.

That’s it. But the editing is where everything happens.

When you edit a comprehensive draft on a topic you’re stretching into, you’re forced into a series of decisions that are each a learning event:

• “Is this actually true?” — You fact-check claims against your existing knowledge and external sources. This surfaces what you know, what you assumed, and what you were wrong about.
• “Does this belong here?” — You cut material that’s tangential. This forces you to identify what’s structurally important versus what’s just related.
• “This doesn’t feel right” — You rewrite passages that are technically correct but miss the point. This is where your own understanding crystallises, because you can only reframe something you’ve internalised.
• “Wait, but what about…” — You add asides, connections, and challenges that the AI draft missed. These editorial additions are your original contribution — the part no synthesis engine can produce, because they come from your specific context, experience, and adjacent knowledge.

The output has two layers: the base text (authoritative, structured, comprehensive) and the editorial layer (asides, annotations, corrections, colour). The editorial layer is the learning made visible. It’s also the thing that makes the writing yours.

Why this works for technical people

Technical people are trained evaluators. Code review, architecture review, PR feedback — the skill of reading something critically and improving it is already in the muscle memory. Learn by Scalpel repurposes that skill for learning.

It also solves the cold-start problem. The hardest part of learning a new domain is generating enough initial structure to have something to react to. AI synthesis gives you that structure immediately. You skip the blank page and go straight to the part where you’re engaging with the material — interrogating it, reshaping it, pushing back on it.

The method works best in stretch territory: topics where you have enough adjacent knowledge to evaluate the draft critically, but not enough depth to have written it yourself. Too far outside your knowledge and you can’t edit meaningfully. Too close and there’s nothing to learn. The sweet spot is material that’s one layer deeper than what you currently understand.

What this is not

It’s not asking AI to write your blog post. The AI draft is raw material, not finished product. An unedited AI synthesis is commodity content — comprehensive, fluent, and interchangeable with what anyone else could generate from the same prompt. The editorial layer is what makes it valuable, because it contains the decisions, judgments, and connections that only you can make.

It’s also not the Feynman method. Feynman says: explain it simply to find your gaps. Learn by Scalpel says: start with a complete explanation and discover what you actually think by deciding what to keep, what to cut, and what to challenge. Feynman works bottom-up from your current understanding. This works top-down from a synthesised overview. They’re complementary, not competing.

Publishing in stages

The web lets you do something print never could: show the work evolving.

Publish early. Publish the AI draft with a visible marker — “Stage 1: Raw synthesis.” Then edit in public. The piece moves through stages, and each stage is labelled so the reader knows what they’re looking at:

• Stage 1 — Raw synthesis. The AI-generated draft, unedited. Present but clearly marked as unreviewed. This is useful to readers immediately — it’s comprehensive, structured, and directionally correct. But the reader knows to hold it lightly.
• Stage 2 — First pass. The author has cut, corrected, and annotated. Editorial asides are appearing. Claims have been checked. The structure may have changed. The piece is becoming trustworthy but isn’t finished.
• Stage 3 — Editor’s cut. The editorial layer is complete. The author’s voice, judgments, and connections are fully present. The base text has been reshaped around what the author actually learned. This is the finished piece.

The stage marker is always visible — a banner, a badge, a header line. The reader never has to guess how seriously to take the content.

This does three things at once:

It removes the publishing bottleneck. You don’t wait until a piece is perfect to share it. A Stage 1 post on a topic someone needs right now is more valuable than a Stage 3 post published three weeks later. People get access to the synthesised material immediately and can watch it improve.

It shows the learning. The progression from Stage 1 to Stage 3 is a visible record of what the author brought to the material. Anyone can generate Stage 1. The distance between Stage 1 and Stage 3 — what was cut, what was added, what was reframed — is the author’s contribution made legible.

It invites collaboration. A Stage 1 or Stage 2 piece is an open invitation: “Here’s where I am. What am I missing? Where am I wrong?” Feedback arrives when it’s most useful — while the thinking is still in progress, not after it’s been published as final.

How to try it

Pick a topic you need to learn that’s just past your current depth. Ask an AI to write a comprehensive, technically grounded explanation. Then open it in your editor and start cutting, annotating, and rewriting. Pay attention to the moments where you disagree, where you want to add context, where something feels wrong but you can’t immediately say why. Those moments are the learning.

Keep your editorial additions visually distinct — marginalia, callout boxes, inline comments, whatever works for you. When you’re done, the ratio of base text to editorial additions tells you how much you learned. A heavily annotated draft means you engaged deeply. A lightly touched one means you either already knew the material or didn’t push hard enough.

If you’re publishing on the web, publish at Stage 1. Label it. Edit in the open. Let people see the cut.