May 29 I wrote my first AI trainwreck story. June 18 I finished #15.
People keep asking if this was some kind of "writing experiment" — it wasn't. I'm not that poetic.
The truth is, nobody was reading what I wrote before.
Scroll back to my early posts: AI agent setup notes, QA automation on a budget, a couple of technical deep-dives. Zero to one reaction each. Comments section was a ghost town. I took a real look around Dev.to — the front page is packed with tutorials and technical walkthroughs, but story-driven content? Barely any.
The old road was a dead end. Time to find a new one.
The results: worst-performing story got 7 reactions, best one got 86. Across 15 stories: 445 reactions, 251 comments.
Wrong direction, corrected.
The series wasn't planned from the start. I just started linking to my other articles at the end of each post so people could hop between them. Then I found out Dev.to has a series feature, and the name followed.
I'm a fan of Love, Death & Robots, so the structure came pretty naturally — AI, Ego & Regret.
Around story #8 or #9, a thought snuck in: could this actually become a book? That's when I set a target of 15 stories.
Wasn't sure I'd make it. Decided to write first, figure it out later.
I also realized the covers needed to be consistent. So I went back and redesigned all 15 covers with the same chessboard background and layout. Now it actually looks like a series.
My Favorite One Did the Worst
#6 — the CTO's 2.8B gateway story. Both sides were smart, the back-and-forth was real, no dumbed-down villain. This was the one I was most proud of. 9 reactions, 2 comments.
Some stories I thought were just "okay" did way better. Couldn't help but laugh at that.
The biggest contrast? #10 — the AI Skill story. It's currently my highest-viewed article, and I still can't quite believe it took off the way it did.
Looking back, I think it hit a nerve that goes deeper than tech: "Your experience gets packaged into a Skill, then you're not needed anymore." That's the sword hanging over everyone's head right now. Not just QA — every engineer.
And #6 — good story, but not scary. It didn't sting. Readers don't want "both sides played well" — they want the villain publicly humiliated. That rarely happens in real life. Adults save face for each other. But in a story, readers want someone to do what they can't.
Numbers in Headlines
I'll be honest — I don't want to be a clickbait writer. But if the headline doesn't grab you, it doesn't matter how good the article is.
The early stories had a clear pattern: big dollar amounts. $500K, $660K, $2.8B, $1.4M, $470K. They worked. Articles with numbers in the title clearly got more clicks.
But the later stories showed me something interesting. #10 had no dollar amount in the title — and it's my most-viewed article. #14 — no number, 55 reactions. #11 — no number, 26 reactions.
A number gets them in the door. What keeps them reading is something else entirely.
The One That Was Different
#14 — the one about AI flagging my article — was the most unusual one in the series.
Not because of the numbers. Because the article itself solved a problem.
At the time, my account had been flagged by the community. New articles were only visible to my followers — they wouldn't show up on the feed.
What pissed me off wasn't the flag itself. I'd been using AI tools, and I'd already gone through all my articles adding AI disclosure where needed. But nobody told me my account was flagged. I published an article I was excited about, and half a day later only a few dozen people had seen it. I had this sinking feeling something was wrong.
I hopped into a thread by Dannwaneri (@dannwaneri) — his post about getting flagged by Sloan was one of the earliest discussions about AI detection on the platform. While we were talking, a community moderator, Francis (@francistrdev), chimed in and told me: yeah, your account's flagged.
The irony? I'd already published the article about AI flagging before that conversation. It wasn't on the feed either.
I wrote that article while my account was still restricted.
Later I went to Francis's Q&A thread and asked three things — how many times does Sloan flag before action, do you get notified, and what do I do when my articles can't be found? He answered every single one. Not an earth-shattering change, but someone actually listened.
Sometimes an article isn't just an article. It's a wrench in the toolbox.
30 Passes
15 stories. Each one went through at least 30 rounds of revision before I'd hit publish.
Logic had to hold up. Numbers had to be precise. No contradictions between chapters.
If it looks fake, I don't publish it. I owe that much to myself and to anyone who reads it.
Not a Single Cup of Coffee
Every single article has a "buy me a coffee" link. All 15 of them.
Zero dollars. Not one.
I don't know if the link's too hidden or if trainwreck-story writers just aren't the type people buy coffee for 😂
But here's the thing: I don't actually care.
The views, the comments, the reactions — that's what actually makes my day. Someone writing "this was great" in the comments. A DM saying a particular paragraph hit close to home. That's worth more than coffee money.
The link's still there, though. Just in case.
So What Was I Even Trying to Say?
15 stories in, I sat back and asked myself — what was the point?
Was it for the attention? Sure, at first. Nobody wants to write into a void.
Was it for the coffee money? 15 articles, zero dollars — that math doesn't work.
Or is it because my own job is getting replaced by AI?
I've been in QA for 15 years. AI testing tools, automation coverage, smart diagnostics — the trainwrecks I wrote about, some I've seen firsthand, some I worry I'll run into soon.
Maybe I was writing these stories to amplify that anxiety. Not just for others, but for myself. A reminder: systems crash, people deflect blame, the 3 AM phone always rings — but the person who picks up that call is worth more than the one who only knows how to run a report.
Every trainwreck story was me telling myself: Your experience cannot be packaged into a Skill.
One Comment I Read 15 Times
Brilliant - I love reading a "David vs Goliath"/"underdog" kind of story like this! Best thing - he didn't panic, and he didn't try to get his "revenge" either (well the revenge was there of course, but purely through his results) ... epic story!
— leob, on the very first story of the series.@leob, if you're reading this — thank you.
That's not empty politeness. Before writing every new story, I go back and re-read this comment.
Not because it was the first nice thing someone said. Because it nails what I was trying to do with every single one of these stories — the protagonist didn't panic, didn't fight back with words, just let the results speak.
All 15 stories, same skeleton, different skin. When I lose the feel for it, I read this comment and know I'm not off track.
Then there's Daniel Balcarek (@gramli), a backend engineer from Czech, who left this on #2:
I really like your writing style. I just read the second story and couldn't stop reading. 😅 Great job!
I've gone back to that one too. "Couldn't stop reading" — three words, and that's all I needed. The best thing a storyteller can hear isn't "your data checks out" or "good technical approach." It's "I couldn't stop."
There are more. @itskondrat (Mykola Kondratiuk) on #3:
what's almost never in a \$660k deployment: explicit constraint on blast radius at that hour. not in the budget, not in the runbook.
One sentence that said the whole thing better than I ever could.
And on #10, two people called me out — Utkarsh Bansal (@utkarshbansal01) said it read like AI wrote it, and framemuse (@framemuse) said it lacked a human touch. I replied to both. One I agreed with, the other I admitted my timeline was messy. Getting called out means someone actually read carefully enough to question it — and engaging with that is way more interesting than ignoring it.
This is what I get paid in. Not coffee. People reading carefully enough to write something back.
How Real Are These Stories?
15 trainwreck stories. They're not my own experiences, but they're not entirely made up either.
I've been in QA for 15 years. How systems crash, how people deflect blame, what a 3 AM phone call sounds like — those are real. I took all of that, broke it down, and reshaped it into characters and situations.
The stories are fiction. The failure modes are not.
What's Next
A Book?
Around story #8 or #9, I realized this might be more than 15 articles. It could be a book.
The plan is to polish all 15 stories, add some new material, and put them together.
But a book is missing one thing: the Praise page.
I don't want to write my own blurb. I want to ask people who actually read the series to write it.
If you read any of these stories — if one made you laugh, or reminded you of something you've been through, or got quoted in a meeting — would you write a sentence or two for the Praise page?
Just a couple of lines. I'll pick a few and put them at the front of the book.
Drop a comment or reach me through my profile email. I'll follow up.
15 trainwreck stories. 445 reactions. 251 comments. 20 days.
See you in the next series.
If you think the series is worth a coffee: ☕
P.S. English isn't my first language. I use AI to polish the writing and help with storycraft. Thanks for reading.
Top comments (27)
After reading all these AI stories, my conclusion is simple: automation can replace tasks, but not context, intuition, and accountability. Thanks for sharing both the successes and the limitations of AI in such an engaging way.
"automation can replace tasks, but not context, intuition, and accountability" — that's the whole series in one sentence. Glad you stuck around through all 15 🙌
@xulingfeng Thanks, cool to read that, much appreciated!
See you in the next series 👊🙌
Sure thing - also good that you explained your main underlying motive for publishing this series - I completely understand and agree, I have the same feeling about some companies' knee-jerk reaction to AI, in that they think it's a good idea to get rid of their employees without much thought "because AI can do all that" (without proper human guidance or oversight) - I think they'll learn their lesson the hard way ...
David vs. Goliath, the score: 15-0 !
That 15-0 scoreboard is yours too. You were there from game 1. 🏆
technical PM writeups I was proud of sat at 2 reactions while a messy 'here's what went sideways' post broke out. story-first isn't just engagement bait - it's the honest format for anything worth learning from.
Exactly this. The messy ones are the only ones worth writing. Appreciate you being here through it all 🙌
yeah, and harder to write too. making sense of the mess is the actual work
That's the part nobody tells you — writing the clean version is easy because you just leave things out. Writing the messy one means you actually have to understand it first.
right, and the things you leave out are the parts that bite you later.
You just described the arc of every AI failure story I've written. The thing that broke was never the thing people were watching — it was the context they trimmed because it seemed irrelevant at the time. Writing and deploying have that in common: what you leave out is usually what kills you.
Which detail bit you hardest that you almost left out?
yeah, the stuff that felt safe to trim always turns out not to be. context is never as stable as it looks.
Interesting!
Consistently consistent. Appreciate you, Benjamin 🙌
that is so true! :)
🤣🤣🤣
Lol :)
Congrats on wrapping up the series 😄
15 stories is no small thing. I can imagine how much work went into writing, revising, and keeping everything consistent.
I'll definitely be checking out whatever comes next.
Thanks Hemapriya! 🙌 The 36 stratagems series starts next — see you there 😄
15 stories.
445 reactions.
251 comments.
The engagement economy is brutal. 😂
Wait, I wrote 15 stories and only got 445 reactions? That's 29.6 reactions per story. I'm clearly not optimizing my engagement-to-effort ratio correctly 😂
This is a useful lesson for technical writing too. The story format works because it gives the reader a concrete failure, a person in the room, and a consequence. A pure AI tutorial often skips that and jumps straight to the tool. The tool is easier to explain, but the failure mode is what people remember.
That last line — "the failure mode is what people remember" — I think that's the whole point of writing in this format. Tutorials age, stories don't.
Reading this made me realize you're one of the reasons I kept writing on dev.to.
When I published my first post a few weeks ago, you were the first person to react, comment, and follow. At the time I had no idea whether anyone would care about what I was writing, and that interaction made it feel like I wasn't just shouting into the void.
I'm only three posts in, but your retrospective made me appreciate how much those early interactions matter. Thanks for being one of the people who showed up at the beginning.
This made my day. I remember that first post of yours — glad I clicked on it. Three posts in and already making an impact. Keep going, man. 🙌
Some comments may only be visible to logged-in visitors. Sign in to view all comments.