What Are the 36 Stratagems?
If you've heard of The Art of War, think of the Thirty-Six Stratagems as its scrappy younger cousin. Sun Tzu wrote about macro strategy — when to fight, when to retreat. The 36 Stratagems are tactics. Specific moves. The kind of play you can see on the board.
A few examples: Deceive the Heavens to Cross the Sea — hide your real intent behind routine actions. Besiege Wei to Rescue Zhao — attack where your enemy is forced to defend. Kill with a Borrowed Knife — use someone else's strength to remove a threat.
They were written for warfare. They work just as well in office politics, courtroom arguments — and, as it turns out, AI projects that crash in spectacular ways.
Both books are on Amazon. I'm not linking them — makes me look like an affiliate marketer, and they're not paying me. 😄
The 6 People
12 years of ops experience was packaged into an AI Skill. Then he was laid off. The CEO's memo: "12 years of experience, now available as a prompt."
Months later, the AI crashed — the company migrated from RabbitMQ to Kafka, but nobody re-validated the Skill. It was still running the old RabbitMQ retry logic. At 4:12 AM, the CTO called him back at 5x his old monthly rate to fix it. He didn't stay. Signed a two-week contract, fixed it, walked out.
His one-person consultancy is still running. One clause: no AI in the delivery chain.
Temperament: Strategist. Thinks three moves ahead. Leaves hidden failsafes in routine operations — like a note in a migration doc saying "450ms is RabbitMQ GC window only. Do not use in other environments."
Yeah, I got to the third story before realizing I never gave them a name. So P it is.
An AI payment gateway processing $2.8B in transactions. The CTO used formal verification to prove it was "mathematically safe." P spent eight months secretly building an adversarial testing pipeline to prove the gateway would approve illegal transactions.
Later, P's own article got flagged as "low quality" by an AI moderation system. P pulled the internal API, grabbed 347 flagged records — the effective accuracy was 38%. False positives outnumbered correct ones.
Temperament: Calm. Data-driven. If you're sitting across from P with a fabricated benchmark, you should be nervous — but P won't give you any signals until the evidence is ready to land.
Open-source AI pilot ran for six months. 91.3% accuracy. Zero incidents. The new CTO spent $8M on his old company's product and fired Leo's entire team of six. On day 13, at 2:47 AM, the $8M system collapsed — six GPUs OOM, three AI agents fighting over context windows. Leo walked back to the server room, brought the old system online. 30 seconds to recovery.
The CEO called at 3 AM: full team rehire, double salary, CTO position. The CEO's exact words — "Your team. Your budget. Your technical direction."
Temperament: High energy. Technically confident. Doesn't play politics — not because he can't. He genuinely believes good code wins in the end. (He's learning the hard way that it doesn't always.)
VeriTest had 15 people. Competitor QualiGuard had 200, claiming 97.2% AI coverage. Lena didn't look at the coverage number — she looked at the test logs. 3 hours 12 minutes. Zero external dependencies. The competitor was running a sandbox.
She paid $5,000 to a former employee for an internal recording. The VP and the test director had budget friction. She fabricated a $1.3M fake quote, leaked it through an intermediary. The VP pressured the director to cut costs. The director swapped full regression for sampling + AI prediction. Real coverage: 86.7%. Lena won.
In the parking lot, the director stopped her. "You planted that quote." Lena's reply: "I never said I'd fight clean."
Temperament: Sharp. Patient. Not afraid to get her hands dirty. Find someone's pressure point, push, and wait. Or buy a recording. Or plant a fake quote. Either way, the board's in her hand.
Backend engineer. He writes things nobody asked him to write.
Marcus Webb demoed a shiny vector database at the all-hands. He glanced at Alex's screen and smirked: "SQLite? For AI memory? Like putting a bicycle on an F1 track."
Over the next six weeks, Alex quietly built a SQLite FTS5 fallback system. Told nobody. At 2:14 AM, the vector DB died — CUDA deadlock. Marcus said it'd take two to three days to fix. The board demo was in seven hours.
Alex typed in Slack: "I have a parallel fallback running. Give me five minutes."
2 minutes 43 seconds. Switched over. First query: 2 milliseconds.
Temperament: Quiet. Paranoid. Doesn't argue when mocked — waits until 2 AM when the system goes down, then cuts over.
QualiGuard's Director of Testing. He walked into the bid review certain he'd already won — 97.2% AI coverage in a sandbox, 4-day delivery, no holes. He didn't know the VP had already cut his GPU budget in half. He didn't know the other side already had the tape of him and the VP at odds.
In the review, Sarah stood up and said three sentences — "The delivery timeline is based on Derek's personal technical assessment, not reviewed by management… If QualiGuard is selected, the revised proposal will be submitted under new technical leadership." Derek sat through all of it. Didn't say a word.
Then he updated his LinkedIn. New title: Director of Testing, medical IT.
Temperament: Arrogant. Stabbed in the back by his own side in a game he thought was already won.
The Framework
Over 36 stories, these six paths cross — some face off, some work together, a few are being moved by forces no one can see yet. And sometimes, they're just solving their own problems, while the reader sees it all before any of them do.
See you in the first story of AI, Ego & the 36 Stratagems.
P.S. English isn't my first language. I use AI to polish the writing and help with storycraft. Thanks for reading. ☕ Buy me a coffee