2026-02-09|8 min read|--kai--mcp--ai automation--content workflow--solopreneur

Training Kai — My AI Assistant That Writes and Publishes for Me

Training Kai — My AI Assistant That Writes and Publishes for Me

## Who is Kai?

Kai isn't a real person. Kai is Claude — but "trained" by me to become a personal assistant.

I gave Kai two knowledge files:

  1. >System Prompt: Role, communication style, principles
  2. >Knowledge Base: Info about me, projects, tech stack, decisions made

Then I set up Model Context Protocol (MCP) — a system that allows Claude to directly interact with:

  • >My blog (Firebase)
  • >Google Sheets (content calendar)
  • >Google Calendar (scheduling)
  • >Cloudflare R2 (upload images)

Result?

I just tell Kai: "Write a post about Hũ Vàng and publish it to the blog."

5 minutes later, the post is live.

## Real workflow: From idea to published post

Yesterday, I told Kai:

"Kai, write a post about Hũ Vàng. The story is my wife bought gold, tracked it with Excel, I found it exhausting so I made an app."

Kai replied:

"Okay. After writing, I'll:

  1. >Publish to blog (gianghaison.me)
  2. >Generate 3 Facebook posts (different hooks)
  3. >Generate 1 Twitter thread
  4. >Generate 1 carousel for Facebook Page
  5. >Save everything to Google Sheet
  6. >Create reminder schedule in Calendar

Do you agree?"

Me: "Agree."

5 minutes later:

✅ Blog post published: gianghaison.me/blog/the-gold-jar-when-my-wife-bought-gold

✅ 3 Facebook posts written, saved in Sheet:

  • >Post 1: Personal story hook (post Monday, 8 PM)
  • >Post 2: Problem/solution hook (post Wednesday, 12 PM)
  • >Post 3: Insight hook (post Friday, 8 PM)

✅ 1 Twitter thread (9 tweets, post Tuesday, 9 PM)

✅ 1 Facebook carousel (5 slides, post Sunday, 8 PM)

✅ Google Calendar has 5 reminder events

I didn't do anything. Kai did everything.

## Kai even knows "content repurposing"

I asked Kai: "How many posts per week is reasonable?"

Kai analyzed:

  • >Blog: 1-2 posts/week (SEO needs consistency, not frequency)
  • >Personal Facebook: 4-5 posts/week (FB algorithm likes "consistent presence")
  • >Twitter: 1-2 tweets/day (Twitter rewards frequency)
  • >Facebook Page: 3-4 posts/week (organic reach is low, don't waste time)

But Kai didn't tell me to write 15 unique content pieces per week.

Instead, Kai proposed: Option C — Repurpose-focused

"Write 1 long blog post (1500-2000 words). I'll break it down into:

  • >3 Facebook posts (different excerpts, different hooks)
  • >1 Twitter thread (5-7 tweets)
  • >1 infographic carousel for Page

1 post → 15+ pieces of content."

I don't need to write 15 posts. I just need to write 1 quality post.

Kai handles the rest.

## Why this matters

Because I'm a solopreneur.

I don't have a marketing team. No copywriter. No content manager.

Just me.

And if every week I had to:

  • >Write 15 unique content pieces
  • >Log into 5 different platforms
  • >Remember to post at the right time on the right day
  • >Reformat content for each platform

I'd burn out after 2 weeks.

But with Kai + MCP:

I just need to focus on 1 thing: Write 1 good blog post per week.

The rest — repurpose, schedule, remind — Kai handles.

## Kai understands my writing style

I didn't just give Kai technical specs. I also gave it my writing principles:

From my knowledge file:

"Sơn's writing style: good storytelling, long sentences with slow rhythm, keen observation, natural humor, light philosophy, conclusions that leave an impression."

So when Kai writes for me, it doesn't sound like "AI-generated content."

It sounds like me.

Because I trained it on:

  • >My previous blog posts
  • >My communication style
  • >What I care about
  • >What I avoid

Kai becomes an extension of me, not a replacement.

## The planning conversation

When I asked Kai about posting frequency, it didn't just give me a number.

It analyzed:

  • >FB algorithm (rewards consistent presence over high frequency)
  • >Twitter dynamics (rewards frequency + engagement)
  • >Blog SEO (quality + consistency > spam)
  • >My capacity as a solopreneur (sustainable > aggressive)

Then it gave me 3 options with trade-offs:

Option A: Conservative (12 posts/week, sustainable)
Option B: Aggressive (17 posts/week, growth-hacking)
Option C: Repurpose-focused (1 blog → 15+ pieces, smart)

And recommended Option C with reasoning:

  • >"Solopreneurs don't have time to write 17 unique posts/week"
  • >"1 quality blog post = content source for the whole week"
  • >"Each platform needs different format, not different content"
  • >"Sustainable, won't burn out"

That's not just an AI assistant. That's a strategic partner.

## The content library system

Everything Kai writes gets saved to a Google Sheet called "Content Library."

Columns:

  • >Date (when to post)
  • >Platform (facebook, twitter, blog, etc.)
  • >Title (post headline)
  • >Content (full text)
  • >Status (pending / posted)
  • >Tags
  • >URL (after posting)
  • >Slug

Then Kai creates calendar reminders:

  • >Monday 8 PM: Post FB #1
  • >Tuesday 9 PM: Post Twitter thread
  • >Wednesday 12 PM: Post FB #2
  • >Etc.

I get notifications on my phone. Open the sheet. Copy content. Post.

Eventually, even this will be automated (MCP v3 with auto-posting).

But for now, manual posting lets me review before going live.

## What I learned from building this system

### 1. AI works best with clear context

The more specific my knowledge files, the better Kai performs.

Bad: "Write a blog post about my app"
Good: "Write a blog post about Hũ Vàng — origin story, why I built it, the 'gold jar' mental model, target users"

### 2. Automation ≠ removing humans

I still write the core content. But Kai removes the busywork:

  • >Reformatting for platforms
  • >Scheduling
  • >Reminders
  • >Repurposing

I stay creative. Kai handles logistics.

### 3. Build systems > hire people (at early stage)

As a solopreneur, I can't afford a content manager.

But I can build a system that does 80% of what a content manager would do.

And it costs $0/month (using Claude's existing subscription).

### 4. The best tool is the one you build for yourself

There are content management tools. There are scheduling tools. There are repurposing tools.

But none of them fit my exact workflow.

My MCP system does — because I designed it for me.

## Real examples

### Example 1: Quick blog publish

Me: "Kai, publish the Hũ Vàng post we just wrote."

Kai:

[Generating slug from title...]
[Uploading to Firebase...]
[Setting status: published...]
✅ Published: gianghaison.me/blog/the-gold-jar-when-my-wife-bought-gold

### Example 2: Content repurposing

Me: "Kai, take this blog post and create Facebook content for the week."

Kai:

Analyzing blog post...
Extracting 3 key narratives:
1. Personal story (wife + Excel)
2. Problem (existing apps too complex)
3. Solution (gold jar mental model)

Generated:
- 3 Facebook posts (different hooks)
- Saved to Content Library
- Created Calendar reminders

Would you like me to generate Twitter thread as well?

### Example 3: Upload artwork

Me: "Kai, upload my Saigon skyline sketch to the gallery."

[I attach the image]

Kai:

[Uploading to Cloudflare R2...]
[Creating artwork entry in Firestore...]
✅ Done. 
Title: Saigon Skyline
Medium: pencil sketch
URL: gianghaison.me/gallery/saigon-skyline

## What's next: MCP v3

Current system requires manual posting. But I'm researching auto-posting:

MCP v3 goals:

  • >Facebook Graph API integration (auto-post to personal + page)
  • >Twitter API v2 integration (auto-tweet + threads)
  • >Scheduled posting (set it and forget it)
  • >Analytics tracking (views, engagement, clicks)
  • >A/B testing (test different hooks, track which performs better)

Ultimate vision:

I write 1 blog post on Monday morning.

By Monday evening:

  • >Post is live on blog
  • >5 social posts scheduled for the week
  • >Calendar reminders set
  • >Analytics tracking enabled

By next Monday:

  • >Kai sends me performance report
  • >Suggests which hook performed best
  • >Recommends posting times based on engagement data

I just create. Kai distributes, schedules, tracks, and optimizes.

## For indie makers and solopreneurs

If you're building alone, you know the struggle:

  • >Too many hats to wear
  • >Too little time
  • >Too much manual work

You can't afford a team. But you can build systems.

My advice:

1. Start with repetitive tasks
What do you do every week that feels like busywork?
That's your first automation target.

2. Use AI as your co-pilot, not your replacement
You stay strategic. AI handles execution.

3. Build incrementally
I didn't build MCP v2 in one day. I started with blog posting. Then added gallery. Then calendar. Then repurposing.

Each piece added value. Each piece was usable immediately.

4. Document everything
My knowledge files make Kai effective. Without them, Claude is just generic AI.

With context, it becomes Kai — my assistant who knows me, my projects, my style.

## Final thoughts

Two years ago, I wouldn't believe this was possible.

An AI assistant that:

  • >Writes in my voice
  • >Publishes to my blog
  • >Repurposes content for multiple platforms
  • >Schedules reminders
  • >Tracks my projects
  • >Adapts to my workflow

All accessible through a chat interface.

No coding required.

Just clear communication about what I want.

If you're an indie maker, solopreneur, or content creator drowning in busywork:

Build a Kai.

Train your AI.

Let it amplify you.

You don't need to know how to code.

You just need to know what you want.

And communicate it clearly.

The rest is automation.

P.S: Yes, Kai helped write this too. But the ideas, the structure, the stories — those are mine. Kai just helped me articulate them faster.