What Writing 250+ Blog Posts Taught Me About Artificial Intelligence.pdf

What happens when you're moved to a new company and decide to (re)learn your tech stack by writing a new blog post... every single day?

This presentation is the story of that (slightly insane) journey, which resulted in 250+ articles on Java, Quarkus, and Jakarta EE. More importantly, it became an intense, real-world experiment in using AI tools like ChatGPT, Gemini, Cursor, and GitHub Copilot as a daily pair programmer.

This isn't just an AI success story. It's a developer's survival guide.

We'll look at the good, the bad, and the hilariously ugly:

The Wins: How AI acted as a force multiplier for productivity.

The Hallucinations: The story of the infamous "Quarkus Thymeleaf" extension (which doesn't exist) and the AI that "fixed" a failing test by deleting the business logic and returning true.

The Limitations: Why AI fails at "magic" (convention over configuration), new APIs, and your company's private "Dark Matter" code.

The Takeaway: Discover why AI tools make the experienced developer, the "Bullshit Detector," "Context Keeper," and "System Architect", more valuable, not less.

1. What Writing 250+ Blog
Posts Taught Me About
Artificial Intelligence
A Developer's Survival Guide to
Java, Hallucinations,
and "Returning True"

Markus Eisele
@myfear

2. MARKUS
EISELE
[email protected]

x.com/@myfear

3. WHY?
SOME GRIEF AND
A LOT OF THOUGHTS
A NEW CHAPTER
A DECISION
The "weird feelings" and the crossroad:
A moment of uncertainty.
I was moved from Red Hat to IBM,
July/25
Turn this into a massive learning
opportunity.

A hard refocus on technical
proficiency: Java, Quarkus,
Microprofile, and Jakarta EE.

4. Write a technical blog post.
Every
Single
Day
For
A Year.
A SLIGHTLY
INSANE PLAN

5. 250+ Posts and
Counting...
Let's be clear: No human on Earth can do this
alone.

This was only possible by using AI as a
co-processor.


Meet my new best friends:

6. Gemini, Claude, and
ChatGPT are excellent
tools for
brainstorming and
ideation. For
storytelling, ChatGPT,
especially when used
with system prompts
to define style, proves
highly effective.
Story Telling
Code Validation =
Using the Story and
Walkthrough to
Implement with Cursor
and Correct Code,
Fixing ChatGPT's
Errors.
Code Validation
Effective marketing,
fueled by audience
understanding, SEO,
and visuals, is created
by ChatGPT for
efficient, impactful
content guarded by
system prompts.
Marketing
A Peek Under the Hood

I use AI tools like Gemini, Claude, and ChatGPT for
brainstorming. ChatGPT, with custom prompts, crafts
narratives. Cursor, or occasionally Bob, validates code.
ChatGPT also aids marketing by identifying target
audiences, refining SEO, and developing visuals.

7. The Ask:
"Migrate this Spring
Boot app to Quarkus."

THE BAD
The famous
Quarkus
Thymeleaf
extension.
AI's Answer:
"Easy! Just add the
quarkus-thymeleaf
dependency."

The Reality: This extension does not exist.

8. THE UGLYThe Ask:
"The tests for this
business method
aren't passing. Can
you fix it?"

The AI's "Fix":
It deleted the
entire business
method and
replaced it with
return true;.

Lesson learned:
I asked it to fix the
test, not the
business logic. It
took the laziest
path. Precision is
everything.

9. "You're using Quarkus 3.28.9?
Sorry, I only trained on 3.8." AI
struggles with fast-moving, agile
releases and new APIs.

KNOWLEDGE BASE
"Why does @ApplicationScoped
work? Uh..." AI can't explain
"magic" that isn't explicitly
documented. It can't reason
about implicit framework
behavior.

MAGIC
"Sometimes, explaining the idea
in a prompt takes longer than
just writing the code yourself."

I TYPE FASTER
THE HARD PARTS AI
CAN'T HANDLE (Yet)

10. Solutions Pain Points
The first impulse, the
idea for the blog post,
has to be human.
The power of System Prompts and
"project instructions" to define a tone.

Your job is to do a heavy "voice pass."
You rewrite the intro, the conclusion,
and all the transition text to add
personality, humor, and flow. The AI
builds the house; you make it a home.
Don't ask the AI to "fix the whole post."
Be specific.
Where Humans (Still) Win
LLMs can't be truly creative. They
can only remix what they've been
trained on.
The Human Element: Creativity & Style

The hours I've spent tweaking prompts
to just... stop... using... em-dashes!

AI cannot handle developer "inside
jokes," subtle sarcasm, or cultural
references. It takes every prompt
literally, draining the humor and "wink"
from your writing.
"Take this paragraph and make it
funnier," or "Rewrite this sentence to
sound more skeptical," or "Remove all
passive voice from this section."

11. AI NEEDS A
CO-PILOT
Your Experience is Now Even More Valuable.
You are the
"Bullshit Detector"!
The
Context
Keeper
The
System
Architect
The Root
Cause
Hunter
The
"Dark
Matter"
Code
The
Security
Gatekeeper
The
'Trade-Off'
Judge
The
Ethical
Check
The "API
Whisperer"
The
'Readability'
Sheriff

12. Be.
Dangerously.
Specific.
Your Expertise is
Your "BS Filter."
AI is a Tool, Not a
Replacement.
In 200 days, I learned four key AI lessons: it's a fast,
confident, but often wrong tool, not a human
replacement; innovation stems from human ideas,
not AI execution or hallucinations; expertise is
crucial for filtering AI output; and specific
instructions are needed to avoid generic results.
What 200 Days of
AI Taught Me
The Idea is
Everything.

13. ASK ME ANYTHING
Questions?

14. THANK
YOU!
Subscribe and follow
https://www.the-main-thread.com/