Tag: Steven johnson

I’ve been thinking a lot lately about how we move past the novelty phase of AI—beyond just asking a chatbot to “write a poem about a turkey” or summarize a meeting—and into actual thinking with these tools.

As a lifelong learner, I’m always on the hunt for workflows that help me synthesize information better. Most of the “AI for writing” advice I see online is pretty generic. But I recently came across a breakdown of how four high-profile writers are making effective use of tools like NotebookLM and Claude in ways that are much more sophisticated than simple text generation.

What jumped out at me is that none of these writers use AI to write for them. They use it to structure, challenge, and code.

Here are the four models that caught my eye.

1. The Triangulated Research Base (Steven Johnson)

Steven Johnson (Where Good Ideas Come From) has a workflow that solves a problem I face constantly: the messy “research phase.”

Instead of treating the AI as an oracle, he treats it as a connection engine. He creates a dedicated notebook (using Google’s NotebookLM) and uploads three distinct types of sources: a primary source (like a raw PDF or study), a secondary source (like a context article), and a multimedia transcript.

Then, rather than asking for a summary, he asks the AI to find the friction between them: “What themes appear in the interview transcript that contradict the historical account in the PDF?”

It’s less about getting an answer and more about finding the blind spots in your own reading.

2. The Diagnostic Editor (Kenny Kane)

This one really resonated with me because it mirrors the experiment I tried recently with my “Bubble Bath” post.

Kenny Kane uses Claude not to generate prose, but to act as a ruthless developmental editor. He uploads a messy draft and runs a “Diagnostic” prompt. He doesn’t ask “fix this,” he asks: “Where does the argument drift? Where does the energy drop?”

He even has the AI analyze his best writing to identify his specific “DNA” (sentence length, vocabulary choice) and then asks it to apply that same tone to his rougher sections. It’s using the AI as a mirror rather than a ghostwriter.

3. The Memo-to-Demo Shift (Dan Shipper)

Dan Shipper at Every is doing something fascinating that changes the definition of writing altogether. He argues that in the AI age, we shouldn’t just describe a concept; we should build a small app to demonstrate it.

If he’s writing about “Spaced Repetition,” he doesn’t just explain the theory. He asks Claude’s Artifacts feature to “Write a React component that lets a user test spaced repetition live in the browser,” and then embeds that little app directly into the essay. The writing becomes 50% prose and 50% software.

4. The Co-Intelligence Loop (Ethan Mollick)

Ethan Mollick focuses on breaking the echo chamber. Before he publishes, he spins up simulated personas—a skeptical VC, a confused novice, an expert in a tangential field—and asks them to critique his draft from their specific viewpoints.

It’s effectively a focus group of one.

---

How to Get Started

If you’re like me, seeing all these workflows might feel a bit overwhelming. My advice? Don’t try to overhaul your entire writing process overnight. Just pick one experiment to try this week.

Here are two simple entry points:

Experiment A: The “Blind Spot” Check (For Research)

If you are reading up on a topic, don’t just take notes. Open Google NotebookLM, create a new notebook, and upload your sources (PDFs, URLs, or pasted text). Then, ask this specific question:

“Based strictly on these sources, what is the strongest argument against my current thinking? What connection between Source A and Source B am I missing?”

Experiment B: The “Ruthless Editor” (For Writing)

If you have a rough draft sitting on your hard drive, copy it into Claude or ChatGPT and use this prompt (adapted from Kenny Kane’s workflow) before you do any manual editing:

“Act as a senior editor. Do not rewrite this text. Instead, analyze my draft and tell me: 1) Where does the argument lose energy? 2) Does the opening hook successfully promise what the conclusion delivers? Be critical.”

I’ve found that using the tools this way—as a partner for thinking rather than just generating—is where the real magic happens.

Which one will you try first?