Most of my articles don't start in front of a blank page.

They start anywhere.

In the street. On my sofa. Walking. At a café table. Just after a meeting. Sometimes at a moment when I thought I was doing something else.

An idea comes back. It insists. It's not an article yet. Not even an outline, necessarily. It is a ball of mixed-up things: a hunch, an irritation, an example, something I remember from work, a sentence that keeps returning, a contradiction I can't yet put into words.

I used to tell myself: I'll write it later.

Then later came, and the idea had lost its force. The subject was still there, but the momentum had gone. The exact example had vanished. The sentence that held it all together had evaporated.

Now I do it differently.

I talk to my phone.

I empty my head

The first move is not to write.

The first move is to empty my head.

When an article idea shows up, I don't immediately go looking for:

• the right title;
• the right outline;
• the right introduction;
• the right phrasing;
• the right conclusion.

I talk.

I talk through what I have understood, what I have been through, what annoys me, what strikes me as interesting. I can repeat myself. I can start at the end. I can mix two ideas together. I can say a wobbly sentence and then fix it thirty seconds later.

That's fine.

At that moment, I'm not writing anything up.

I'm getting the raw material out.

That's the whole point of the setup: not to ask thinking to be clean too early. An article needs a structure, a progression, a certain discipline. But the idea itself usually turns up in a far rawer state.

The phone lets me respect that moment.

I can talk without putting anything in order.

The phone is wired into the real workspace

The phone here isn't just a voice recorder.

That matters.

When I talk, what I say doesn't end up in an isolated note I'll have to track down later. My phone is wired into my computer. My computer is wired into an AI that can work in my files. And what I say is added straight to the session source of the right Context.

The Mac works. Codex runs. The AI reads and writes in the right folder.

Meanwhile, I can be somewhere else.

That's what makes the workflow different from ordinary dictation. Dictation captures what you say. My system brings what I say into a workspace that already has a purpose, sources, notes, drafts, sometimes decisions or earlier texts.

The capture isn't only a backup.

It already places the idea in its setting.

This is not just swapping the keyboard for the voice

You might think this workflow amounts to using a word processor on a phone.

That would miss the point.

A word processor takes in what I write. It can be very good. It can sync my files, transcribe my voice, let me fix things and move paragraphs around.

But it stays passive.

It doesn't ask me what's missing.

It can't tell that an idea is fragile.

It doesn't notice that I've forgotten the central example.

It doesn't remind me that I've already written something close to this.

It doesn't turn a spoken hunch into a working question.

With the phone wired into the Context, what I say isn't merely typed up. It immediately becomes raw material the AI can interrogate.

The text isn't just sitting there waiting to be reread.

Work on it has already begun.

Nor is it AI at the desk

I could do much of this work in front of my computer.

Open the AI. Explain the idea. Ask for objections. Build an outline. Write it up.

I do that too.

But it's not the same moment.

Sitting at the desk assumes I am already settled into a work session. You have to be in front of the screen, available, ready to organise your thinking.

Plenty of ideas aren't born in that state.

They turn up while the body is busy with something else:

• walking;
• on public transport;
• coming out of a call;
• at a café table;
• in the evening, when the subject keeps turning over;
• at a moment when I had no plans to work.

If I wait until I'm back in front of the computer, I usually keep the subject, but I lose the temperature of the idea.

The temperature is the energy of the moment: the exact example, the connection that has just appeared, the reason the subject matters to me right now.

The phone lets me take the idea while it is hot.

The Context keeps it from staying an unsorted pile that gets lost.

The AI lets me start questioning it straight away.

Most articles get out fast

I also work this way because I want most of my articles to get out fast.

Not in six weeks.

Not after building some heavy setup around the subject.

In my case, they have to be able to get out in two days at most.

That doesn't mean everything is always published within two days, nor that the first draft is already the finished article.

It means the essentials have to get out of my head very quickly.

I can empty out:

• what I already know;
• what I have just understood;
• the examples coming back to me;
• the objections I can feel;
• the limits I want to keep;
• the sentences that carry the hunch.

That short span matters.

It gives the workflow its rhythm. I don't need to build a large body of documentation straight away. I usually start from a recent experience, a problem I ran into, a conversation, a change of mind.

The raw material is still in my head.

It has to get out before it settles.

The session source keeps it raw

When I talk like this, I'm not producing a clean note.

I'm producing a session source directly.

It's raw material: spoken, disordered, sometimes repetitive, sometimes incomplete. It keeps what I say, but also what I clarify, correct, accept or refuse in the exchange with the AI.

This session source is allowed to be messy.

It even has to be, a little.

If it gets too clean too early, it loses part of what makes it useful. It starts to look like a draft, when it should be the primary raw material the draft is made from.

A published article has to be clear.

The session source has to be faithful to the moment the thinking came out.

It keeps things raw so the work can begin.

The AI brings the gaps to the surface

Once the raw material is down, I don't immediately ask the AI to write the article.

That would be too quick.

First I ask it what is missing.

That's often the most useful moment.

When I write on my own, a lot of the gaps show up later. I reread the text two days on. I go for a walk. I take a shower. An objection surfaces. I suddenly see that an example is missing, that an idea moves too fast, that a distinction isn't clear.

That moment is precious.

But it often comes late.

With the AI, it can come in ten seconds.

It can bring up:

• a blind spot;
• a contradiction;
• a missing example;
• a blurry distinction;
• an unexplained assumption;
• an idea too weak to stand on its own;
• a tempting line of thought that's off topic.

And inside a Context, that feedback is more interesting than a plain list of generic objections.

The AI can compare my raw material against what already exists:

• earlier texts;
• the notes;
• the session sources;
• the decisions made;
• the angles already ruled out;
• the purpose of the article.

It can see that a passage repeats another article, that an example doesn't really support the claim, that a conclusion goes further than what I actually said.

It doesn't replace thinking.

It makes the gaps appear faster.

Only then do I build the text

Next comes the outline.

The outline isn't the starting point.

It comes after the raw material and the first objections. That's why it's better. It's not trying to organise emptiness. It organises something that has already started to live.

I ask for a structure. Then I correct it. I move a section. I cut one. I add another. I say the article has to open on a scene, or on a claim, or on a contradiction.

Two or three rounds are usually enough.

At that stage, the article almost exists.

Not in its sentences yet.

But in its logic.

When I finally ask for a write-up, the AI isn't starting from nothing. The ideas are there. The examples are there. The objections are there. The decisions are there. The outline is there.

The AI turns recent, disordered, spoken thinking into readable text.

That's an enormous amount of work. When you talk on and on, you repeat yourself, you double back, you mix two ideas together, you sometimes give an example before you've set out the problem.

Sorting all that out by hand is a chore.

The AI groups, cuts, rephrases, smooths, picks the thread back up.

I see no reason to deny myself that.

Why it's still my article

The question comes up often: if AI does the writing, is the text still mine?

For me, the answer depends on what you hand over.

If I hand over the idea, the angle, the experience, the examples, the judgement and the responsibility, then no, it's not really my article any more.

But if the AI turns my raw material, my reasoning, my choices and my corrections into readable text, then the article stays mine.

What counts is not that every sentence came straight out of my fingers onto a keyboard.

What counts is that the ideas, the organisation, the nuances, the angle and the choices come from me.

After the write-up, I reread. I put a nuance back in. I cut a sentence that sounds too sure of itself. I add an example. I refuse a sequence that rings false.

The AI speeds up the transformation. It doesn't make the calls for me.

What this workflow changes

This workflow has changed my relationship with writing.

Before, writing usually meant finding a free moment, a desk, a keyboard, enough concentration, and then starting again more or less from scratch.

Now the article can start when the thinking arrives.

The real change comes from the combination:

• the phone captures at the right moment;
• the computer gives access to the real workspace;
• the Context files what I say into a session source;
• the AI brings the gaps to the surface;
• the dialogue clarifies the idea;
• the outline sets the progression;
• the write-up turns it into text;
• the rereading makes the text properly mine.

Taken separately, every one of these already exists.

Together, they change the nature of the work.

It's not just writing faster.

It's not just dictating.

It's not just chatting with an AI.

It's bringing a thought that's still warm into a system that can keep it, question it and turn it into something else.

And for an article, that's often exactly what I need.