1/6 done. 30 to go.
After shipping #6, I did the math — 6 stories out of 36. Exactly one-sixth. I have a habit of pausing at every milestone to write something down, otherwise I forget.
When the old series ended, I wrote a recap called "15 AI Stories Later, Some Honest Words." This one follows the same rule — say what worked, say what didn't. No point lying to myself.
Why I Switched Series
Somewhere around story #14 of the "AI, Ego & Regret" series, I started feeling something.
Not that the 15 stories were bad — I don't regret writing them. Some had villains who actually fought back. Some had protagonists who were pawns and didn't know it. Some wins weren't even technical — they were political plays. Every single one, on its own, I'd stand by.
But around #14, I caught myself asking a different question. Not "what's the next story about" — but "why does every story run on the same engine?"
Every win followed the same structure: System breaks → protagonist gets called in to fix it → someone finally sees their value. No matter how good the journey was, the engine never changed. It wasn't about losing. Each story ended with a win. It was about repetition — same structure, 15 times in a row. You win. You also get bored.
Then I wrote #15. Lena's first appearance.
She was different. She wasn't waiting to be rescued — she had cards in her hand from the start. Writing her felt completely different. It wasn't about turning things around. It was about who moved first.
I only understood the difference later: the old series was "waiting to be recognized." The 36 Stratagems is "setting up the board yourself."
Before #15, my characters were all waiting for the system to prove they were right. Lena didn't need the system to prove anything. She was the system. Not a technical duel — a mind game. She wasn't there to fix things. She was there to win.
After that story, I knew I couldn't go back.
So I started this series. 36 Stratagems — six protagonists, each using one stratagem in their first appearance. They all won. But later, when they cross paths, someone wins and someone loses.
Whether they know which stratagem they're using or not — I know. And so do the readers.
Six Boards
Mark (#1 Deceive the Heavens to Cross the Sea) — He wasn't there to read the report. He saw 44 matching entries from a public dataset and a sticker on Torres's computer. The VC hadn't even finished talking before Mark had already placed his pieces.
Derek (#2 Besiege Wei to Rescue Zhao) — He promised an impossible 95% accuracy target. Not out of naivety — he was trading that number for 18 days of peace. He knew exactly what he was gambling against.
Lena (#3 Kill with a Borrowed Knife) — Three knives placed in advance. The ops director went after the budget. The security architect went after the compliance red line. The demo let the data speak for itself. She never raised a finger.
P (#4 Wait at Leisure While the Enemy Labors) — P saw through both vendors in the first week. But P didn't speak — because if you know too early, no one believes you. P waited three months for the data to become its own evidence.
Leo (#5 Loot a Burning House) — An anonymous message at 1 AM. One glance at the failure pattern and he knew what was wrong. Not because he was smart — because he'd seen the exact same trap five years ago. While everyone else was still looking for the root cause, he was already driving out the door.
Alex (#6 Make a Sound in the East, Strike in the West) — A yellow I/O alert. He used it to trigger a full investigation. Everyone watched the dashboard. He watched the pipeline. By the time the team finished checking I/O, he'd already changed the config from 0.7 to 0.0.
The numbers:
| Post | Lead | Stratagem | Views | Reactions | Comments | Read |
|---|---|---|---|---|---|---|
| Teaser | — | — | 192 | 40 | 19 | 4.3min |
| #1 | Mark | Deceive the Heavens | 257 | 56 | 31 | 3.9min |
| #2 | Derek | Besiege Wei | 97 | 36 | 8 | 2.1min |
| #3 | Lena | Borrowed Knife | 86 | 45 | 14 | 1.8min |
| #4 | P | Wait at Leisure | 171 | 47 | 36 | 2.6min |
| #5 | Leo | Loot a Burning House | 230 | 48 | 18 | 1.6min |
| #6 | Alex | Sound East, Strike West | 77 | 23 | 17 | 2.1min |
Total: 1,110 views / 295 reactions / 143 comments
Honest take: #4 had the most comments — 36 in total, ranging from detection-vs-coverage debates to "if you know too early" to taxonomy mismatch. That one hit people in different ways. #2 was the one I feel worst about — Derek's situation was real, but the numbers didn't catch up. #3 too — Lena's prequel, same story, great piece, quiet launch. Different stories have different rhythms. Not a problem.
The Hardest Part Isn't the Stories Themselves
It's getting six people through 36 steps in the same universe without stepping on each other's toes.
That's the biggest difference from the old series. The old series was beads on a string — each story independent, one line. The 36 Stratagems is a web — six threads running at the same time, knotting at specific points.
The six characters were pulled from the old series. The two series share the same universe. Every character's personality and background was locked in the teaser post. I set my own rules.
The outline went through several versions. Who meets who. Who helps who. Who sets who up. Each version got scrapped. Because you're not trying to write N good stories — you're trying to write N good stories that don't contradict each other in the same world. The first six were easy — solo acts, one stage per person. Then the network starts.
The six stories you've seen? Not the first draft.
The Series Started Writing Me Back
There's something I didn't expect about this series. Before I started, I thought I was "designing" six characters — giving each one a stratagem, deciding where they'd cross paths. I was the chess player. They were the pieces.
Then it shifted.
I started using the same tactics I was writing about in real life.
One concrete thing. There's a senior exec who's exhausting. He loves good-looking reports. So I gave him good-looking reports. What he saw and what I was actually doing — two different things.
I was writing Alex using "Make a Sound in the East, Strike in the West" — a yellow alert to pull the team toward I/O, while he changed the config at 2 AM. Then I caught myself doing the same move in real life.
Kinda funny. You think you're writing fiction. Then the fiction starts writing you.
Not a coincidence. I realized later — I couldn't have made up those six characters' moves out of thin air. You can't write the feel of a move you've never pulled yourself. Lena's borrowed knives, P's patient waiting, Alex's misdirection — I didn't invent those. I've seen them, used them, or been burned by them.
What happens when the 36 Stratagems is done? Probably relief — but not emptiness. Because the third series is already stirring. Too early to talk about it though.
That's enough on that.
The Comments Section
The first one in was leob. "Intriguing!" Three words. Then he showed up for every single one. On #2: "choose your battles wisely." On #3: "another banger." On #4: "all-you-can-eat." On #5 he recognized Adyen and said the world is small. On #6: "quieter but subtle." He's been here since the very first Regret story. Never missed one.
Divyanshi has been here since the teaser — seven for seven. On #1 she called out the "benchmark trap." On #6 she wrote "absence of alerts is more dangerous than presence" — some lines I didn't realize were that heavy until she read them back to me. Every single one is a long read. Every single one goes to the deepest layer.
mote showed up for four stories. On #1 he dissected the RabbitMQ-to-Kafka benchmark gaming. On #4 he unpacked taxonomy mismatch. On #5 he walked through the topology blind spot. On #6 he explained confidence threshold degradation — clearer than I did in the original piece.
UnitBuilds dropped what he called a "mic drop with four tables" on #4, then left a few hundred words on #6 — from log tagging to his own project V.A.L.I.D. We went back and forth for a few rounds. It's practically becoming its own mini-series.
Mykola Kondratiuk went deep on the teaser and #1. "Smarter agent, same wall" — he summed up what this series is about in one sentence.
Mike Czerwinski wrote a long analysis on #1, unpacking "production noise as integrity signature" better than I could have.
Vinicius Pereira broke down detection rate vs coverage using Goodhart's Law on #4. "Whoever raises it inherits the bad news" — went straight into my notes.
Self-Correcting Systems analyzed Lena on #3 — "refusing to argue on the surface someone else picked for you" — accurate enough to quote.
Natasha0824 left a long comment on #4 saying "if you know too early, no one believes you" helped her make a real-life decision. I read that one a few times.
There are more. Hemapriya Kanagala showed up across three posts, catching the connections between characters. Evans Owusu linked his own product on #2 — got 5 likes, the highest on that post. Kartik N V J K said "misses over hits" on #4. Alex Shev wrote "a monitoring POC without tests is theater."
Every comment, long or short, pushed me to write the next one.
And one more — Gabriel Weidmann. You were the first person to actually buy a coffee. Two cups worth. I stared at the number thinking — one more and it's a third cup. Then it hit me: "The Third Cup" is the name of a café in this series. Some coincidences you can't write.
What's Next
30 more. In order. One stratagem at a time.
From here on, it gets real — six people aren't on separate boards anymore. They start to meet. Help. Set up. Win and lose.
The breadcrumbs are already placed. As for the third series — that's 30 stories away. Too early. But it's not empty.
About the Book
The old series is fully polished. Two things are still stuck: picking reader comments for the opening pages, and the cover design. I've had AI generate a few cover concepts — none quite right. Anyone have cover design ideas? And pricing — never published a book, no idea what's reasonable. If you've been through it, I'd love advice. Paving a path beyond the day job, honestly.
Thanks for reading this far.
P.S. English isn't my first language. I use AI to polish the writing. Appreciate you being here. ☕ Buy me a coffee
Top comments (15)
Huge thanks to everyone who's been reading and following along — this series exists because of your comments and reactions 🙏
@leob @technogamerz @motedb @unitbuilds @itskondrat @jugeni @vinimabreu — you've been here dropping insights on each story, catching details I wasn't sure anyone would notice. Appreciate you all. 👊
Well it is a good series, honestly, it reads like a mystery novel, if it were written by an engineer. One could say you 'engineered' a good story 😁(I'll see myself out)...
But seriously, well done and the attention to detail is incredible, keep it up, I'm loving the series!
"engineered a good story" — I walked right into that one 😂 I'll allow it. But seriously, your comments across #4 and #6 have been some of the best ones on this series. The V.A.L.I.D. mention on #6 stuck with me. Keep em coming 👊
Will do 😁 Might even drop some more teasers in them too, like this:
I took a step back from the OS and decided to polish up the IDE first, it uses rust and nda, NMCP and Cloudflare workers ai, which to save a dime, I set up with account pooling, so currently 17 free accounts are hot-swapped to get a decent amount of work done. After switching up some packages, it's now at a fixed 109.9MB ram utilization (for 24h straight now, while it literally rebuilt the IDE a few times).
109.9MB fixed for 24h through multiple rebuilds? That's not optimization, that's a flex.😂 OS next?
OS is already working, but for it, I'm currently down the rabbit hole of over engineering the quantization method, to see if I cant squeeze in an even smarter model. Cuz Qwen 2.5 coder 0.5b is great, but how cool would gemma 4 26b a4b be as the core of it all? Cuz that would make a pretty damn smart OS, especially if I can get the quantization just right, then it'll use around the same amount of ram as windows, except you get something useful out of it 😁
Hope you noticed the little .nda files in the top left of the screenshot. Those little beauties right there, is how it doesnt forget a single thing, without slowing down, or eating all your ram.
Out of curiosity — what GPU are you running? I'm on a 4070S with 8GB VRAM. Been testing various small models lately, they're fine for basic chat, but pairing them with Hermes agents is another story. That's where it really gets brutal 😂
Well depends... I have 2 PCs (GTX 1070 and RX 9060 xt 8gb) and a laptop, but due to back injury, I'm working from bed on the laptop (MX250 2gb). Though I do multi-task across them for other work (eg. my day-job) through RustDesk (for now, I should probably switch to V.E.L.O.C.I.T.Y. Remote)
That's brutal — working from bed with a back injury. Hope you recover soon man 🙏
Thanks, though not too soon, cuz it's the only thing keeping me from having to go to the office in another town for work 🫠 After a year of remote work, rebuilding about half of their codebase, over 6000 files edited, apparently they want me in the office, because there's a missing synergy... So I'm job hunting 😅 I dont wanna drive an hour there and back everyday with fuel prices the way they are now, I'd be lucky if a break even end of month.
Especially cuz I know exactly why they want me in... And it's not synergy, it's thievery. They want me there, working on their computers, so they can clone my setup... Because even though all of them are running Claude, I still get 10x the work done compared to them because of my foundry. So I'm desperately job hunting 😅
Wait— so their "synergy" is literally wanting to clone your environment? And you're sitting on a remote desktop project called V.E.L.O.C.I.T.Y.? 😂 Sounds like you've already built your own exit strategy. Hope the back heals and the job hunt lands quick.
Pretty much... Was clear when my manager casually asked 'you use a local LLM for your work, right' yesterday. They know I do and I have since the start. Before they even touched AI, I was using it daily, now that they're on Claude and seeing their subscriptions barely handle 3 days of work a week, while I'm still cruising past their output rates, without even touching my company paid claude subscription...
And yes, I do 😅 I also have a secure messaging and file share app. V.E.L.O.C.I.T.Y. is the core framework of practically everything I make outside the day-job, cuz it's fundamentally better at everything. Remote for instance should be capable of getting close to 900 fps streaming, at line-rate latency (sub-ms processing), I had in mind that for gaming tournaments, they can host all the players on a single server and just section off their peripherals, making it cheaper and lower E2E latency. I had to build everything from scratch, because not a single off the shelf package was fast enough.
And more shoutouts —
@kenielzep97 @natasha0824inkf @hemapriya_kanagala @evans_owusu_6801c8d54ae89 @kartik-nvjk @alexshev @gabrielweidmann — your comments on specific stories (the "misses over hits", the POC theater line, the first coffee ☕) genuinely made my week. Thanks for reading this deep. 🙌