How to Use Claude as Your AI Writing Partner: A 2-Step Workflow That Actually Works

Originally inspired by @vista8's writing workflow on X/Twitter

---

Writer's block isn't really about not knowing what to say. It's about the paralyzing gap between having thoughts and getting them onto the page. Every developer who's tried to write technical documentation, a blog post, or even a detailed GitHub README knows this feeling intimately. You have the knowledge. You understand the system. But staring at a blank document feels like staring at an empty terminal with no idea what command to type.

The good news? There's a deceptively simple AI-assisted writing workflow making the rounds in developer circles — shared by @vista8 — that cuts through this paralysis entirely. It uses Claude as a collaborative writing partner, not a ghostwriter, and the distinction matters enormously. Here's how it works, why it works, and how you can adapt it for your own technical writing practice.

---

Step 1: Brain Dump First, Quality Later

The first step sounds almost embarrassingly simple: get everything out of your head without filtering it.

In the original workflow, @vista8 uses Wispr — a voice-to-text tool for macOS — to speak ideas out loud directly into Claude, since Claude's desktop app doesn't yet have a native voice input mode. The idea is to talk freely, the same way you'd explain something to a colleague over coffee. No grammar policing. No logical structure. No worrying about whether your argument flows.

"You don't need to care about grammar, logic, or order. Say whatever comes to mind."

This works because the real enemy of writing isn't writing badly — it's not starting at all.

For developers and AI engineers, this step might look like:

• Dictating your raw thoughts about a new API integration you just shipped
• Speaking through the architecture decisions you made (and the ones you almost made)
• Dumping a half-formed opinion about a framework comparison
• Pasting in old Slack messages, previous drafts, or screenshots of code comments

You can combine voice input with context artifacts. Throw in a screenshot of your architecture diagram. Drop in a previous draft. Copy-paste a relevant Stack Overflow answer you bookmarked. The point is context richness over editorial polish at this stage.

Here's a simple example of what a raw brain dump might look like when transcribed:

okay so basically we switched from REST to GraphQL last quarter
and like... the main reason was the over-fetching problem
our mobile team was downloading like 40 fields when they only needed 5
and latency was killing us on 3G connections in Southeast Asia
we tried field masking but that felt like a band-aid
also the backend team hated maintaining three different endpoint versions
anyway we moved and it was not smooth, the N+1 problem hit us hard
DataLoader saved us but it took two weeks to figure out
would recommend but only if you have buy-in from both frontend and backend

That's not a blog post. That's not even a coherent paragraph. But it's real, and it's yours, and it contains everything you need.

---

Step 2: Let Claude Play Journalist

Once your brain dump is in, resist the temptation to immediately ask Claude to "write a blog post about this." That's where most people shortcut the process and end up with generic AI content that sounds like nobody in particular.

Instead, ask Claude to interview you.

The prompt is refreshingly minimal:

"I want you to act as a journalist interviewing me about this topic.
Ask me questions — one at a time — to help me clarify my thoughts,
fill in gaps, and uncover insights I might not have realized I had.
Here's my raw thinking so far: [paste your brain dump]"

Claude will start asking targeted questions that force you to articulate things you hadn't consciously formed yet:

• "You mentioned latency was the main driver — was there a specific incident that made this urgent?"
• "What would you tell someone who's considering GraphQL but nervous about the learning curve?"
• "You said DataLoader 'saved you' — can you walk me through what the debugging process actually looked like?"

These aren't generic questions. Because Claude has your context, the questions are specific to your situation. And answering them — even briefly — produces the raw material for your most compelling paragraphs.

This journalist mode works particularly well for:

| Content Type | Why Claude's Questions Help | |---|---| | Technical tutorials | Forces you to articulate why, not just how | | Post-mortems / incident reports | Surfaces the timeline and decision points | | Opinion pieces / hot takes | Stress-tests your argument's weak spots | | Case studies | Extracts the narrative arc from raw data | | Documentation | Identifies what a new user actually needs to know first |

After 4–6 exchanges, you'll have dramatically more material than you started with — and it'll be your voice, because you actually said it.

---

Step 3: Let Claude Assemble, Then You Edit

Once the interview is done, you can ask Claude to synthesize everything:

"Based on our conversation, please write a structured first draft of a
[blog post / documentation page / LinkedIn article] targeting [your audience].
Use my language and examples where possible. Keep my voice intact."

The key phrase is "keep my voice intact." Claude is good at this when you give it enough of your own words to work with — which is exactly what Steps 1 and 2 generated.

What you'll get back is a structured draft that actually sounds like you wrote it, because in a meaningful sense, you did. Your job now is editorial: cut what's redundant, sharpen what's vague, add the code snippets or data points that only you have access to.

This is fundamentally different from asking an AI to "write me a blog post about GraphQL." That produces content. This produces your content.

---

Why This Workflow Matters for Developers and AI Engineers

Most AI writing advice treats the model as a replacement for the human writer. This workflow treats it as an amplifier.

The technical writing problems developers face — imposter syndrome about whether their experience is "worth sharing," difficulty translating tacit knowledge into explicit prose, the time cost of writing versus building — are all addressed by this approach:

• Imposter syndrome: The brain dump externalizes what you already know, making it visible
• Tacit knowledge: Claude's questions force you to make the implicit explicit
• Time cost: The interview phase is fast; you're not staring at a blank page, you're having a conversation

It also produces something that AI-generated slop fundamentally cannot: authentic technical perspective. The N+1 problem story above, the specific context of Southeast Asian mobile users, the two weeks spent on DataLoader — those details exist only because a human lived through them. Claude's job is to help surface and structure them, not invent them.

---

Getting Started Today

If you want to try this workflow right now:

1. Open Claude (desktop or web) and paste your brain dump — voice or text
2. Use this starter prompt: "Act as a journalist. Ask me questions one at a time to help me develop this into a polished piece. Start with the most important gap you notice."
3. Answer 5–6 questions conversationally — don't overthink your answers
4. Ask for a first draft with your audience and format specified
5. Edit with your expertise — add the technical details only you know

The workflow won't write your best work for you. It'll help you write it yourself, faster and with less friction than you thought possible.

---

Want more AI automation workflows, Claude prompting techniques, and developer productivity tools? Explore more on ClawList.io — your hub for OpenClaw skills and AI engineering resources.