I went off the radar for 3 weeks to build my own content machine

After years of going dry, I got seriously back into content creation last November:

• I post daily on LinkedIn
• I revived my weekly newsletter
• And I published a few blog articles

Honestly, it felt great. I had missed creating content terribly.

It gave me new ideas. It helped me sharpen my thinking on certain topics. And I reconnected with the community.

But after a few months, I started to feel a bit lost.

I had the nagging feeling that my content was going in every direction at once, with no clear editorial line. I was spending a lot of time on it… but for what, exactly?

I needed to sort all this out. That’s why I went a bit quiet these past few weeks.

So what did I do?

I built my own content creation tool.

It’s the ULTIMATE tool I’d been dreaming of. And since building tools is literally my job, well… I treated myself.

I tried a few existing ones, but none of them manage to blend editorial planning, content organization, and AI assistance.

So I coded my own. With a Shiny app (of course), which I called MarketChamp’.

Concretely, what does it change for me?

Before diving into the details, here’s the operational result:

• Capturing an idea takes me 10 seconds, wherever I am
• Qualifying it takes 5 minutes
• I never again start from a blank page when it’s time to write
• My publishing calendar self-balances by design across my pillars and objectives

That last line was the hardest to nail. I’ll come back to it.

To show you how the machine runs, the best way is to follow a single idea from the moment it pops into my head to the moment it goes live.

Monday 2pm: the idea lands

I’m in a meeting with a client. He tells me he has an internal Shiny project that’s been crawling for 6 months and nobody knows why.

Bing. Post idea.

Before: I would have had this brilliant idea. I would have told myself I should write about it. And I would have gone back to the conversation with the client. Of course, after the call, the idea was completely forgotten.

Now, I open my email app, I send an email to a dedicated address. Subject: “Slow Shiny apps — anecdote client X”. Body: 2-3 lines to remind me of the context.

That’s it.

On the MarketChamp’ side, a job runs continuously and fetches new emails. The idea is stored in the database and waits.

The goal, with email, is to have the lowest possible friction. At pretty much any moment of my life I can send an email.

Friday morning: the qualification session

Once a week, I sit down for 30 minutes in front of MarketChamp’. My raw ideas are waiting for me, and I’m going to turn them into qualified ideas.

What’s a qualified idea?

An idea that contains everything my “future self” will need to write the post without thinking:

• A narrative angle (not the topic, but the direction of the post)
• An emotional tension (what the reader should feel, or what I felt)
• The concrete details to draw on (names, numbers, juicy specifics)
• The open questions to settle when writing
• A strategic anchor: pillar + vertical + objective

But doing this work by hand for every single idea is hard. “Uhh… the emotional tension?”

This is where AI enters the stage.

AI as a co-editor

AI can’t write in my place.

But it can help me in my creative process. Turning a raw idea into a qualified idea basically means synthesizing and organizing a verbal mush (what I have in my head) into a usable structure. And LLMs are very good at that.

Concretely: I click on the “Slow Shiny apps” idea. A sidebar opens with a chatbot. I tell it everything I have in mind, in bulk. The client context, what struck me, why I think it’s worth telling, the technical details.

The more I say, the better.

It then suggests:

• A polished title
• A narrative angle (“start from the client’s complaint to debunk the myth of slow code”)
• A tension (“frustration of a team that doesn’t know what to try anymore”)
• A list of concrete details to keep handy
• A coherent strategic anchor (R/Shiny Expertise → performance → Awareness objective)
• And it fills in the form directly via tool calls

The AI doesn’t invent anything. It only shapes the raw material I gave it.

I review, I adjust, I validate. It took me 5 minutes.

And the calendar?

That’s the other thing the chatbot does for me. My editorial calendar is cyclic over 8 weeks. Each day of the cycle is pre-assigned to a pillar and an objective. After 8 weeks, the cycle starts over.

Why this system? Because it forces variety by design. I no longer have to ask myself “am I talking about Shiny too much lately?”: the answer is in the calendar.

Once the idea is qualified, the chatbot calls a find_available_slots tool that looks for the next free slot matching the idea’s pillar+objective. It suggests a date. I validate. It’s scheduled.

The idea moves to planned status.

The following Wednesday, 9am: writing

On D-day, I sit down in front of MarketChamp’’s editor. On the left: the qualified card for today’s idea. On the right: a text editor.

In front of me:

• The title, angle, tension, details
• The pillar and objective
• The posts from the same week (to manage LinkedIn ↔ newsletter synergy)
• My personal style guide
• An assistant that can draft a first version if I ask for it

The post writes itself in 20 minutes. Because all the thinking work was done upstream, during qualification. I just have to “tell the story”.

The idea moves to drafted, then done once published.

What works, what flops

Originally, what I really wanted was to invent my methodology, and build a tool around that methodology. I did not anticipate at all that AI would end up playing such a big role.

For me, as of today, AI in content creation = useless. It’s really the textbook case of “AI slop” with generic, incoherent, boring content.

By the way, I added a “Generate a draft” feature in the writing phase. And even when I feed it several months of content, it’s still bad. Sometimes it helps me get past the blank page a little, but I always have to rewrite it myself.

But on the qualification phase, I’m blown away. In one session, I can plan several weeks of content. And above all, the agent is autonomous when it comes to planning. It finds available slots, can swap topics around, checks whether the sequence of subjects is coherent, etc.

It’s a real assistant.

Where there’s still a lot of work to do is in the app’s UX. I built this on my own, on the fly, and the Shiny UI pieces are stacked on top of each other.

It works… for me… most of the time.

But it remains a living tool. The moment I feel the need, I add the feature I need.

That’s the upside of coding your own tool.

By the way, earlier today I added the module to qualify, and write, my newsletters. See you on the other side :)

• Charles