To dispose of an enemy, make use of another enemy. Use a second party to deliver the blow yourself. Bide your time.
— The 36 Stratagems, "Kill with a Borrowed Knife"
Lena never said no. She didn't have to.
She hadn't learned to play in the open yet — because she didn't have that luxury.
97.2% Coverage, 4-Day Delivery, and a Parking Lot That Changed Everything
Some readers already know how Lena made her first appearance —
Lena and her 15-person team at VeriTest won a $1.8M contract against a 200-person AI company with a fake quote and a footnote that said "External Deps: 0." She didn't throw the punch herself at the review meeting — she let the competitor's boss cut their own budget, and let the client's risk officer ask the one question that mattered.
In the parking lot afterward, she said one thing to Derek Shaw, who had just lost the bid: "Next time you put together a proposal, make sure your boss knows what you're doing out there."
That was Lena after she'd learned how to play the game.
This story is about before she learned.
Take Another Look
NexPay was a cross-border payments fintech. Not big, not small. At the Q3 tech all-hands, CTO Vince dropped a bombshell — the company was bringing in DeepCover AI, a platform that claimed to be "AI-driven fully automated testing."
Vince put it this way: "DeepCover AI covers 95%+ test coverage, 10x faster than our manual testing. Already validated at 3 fintech clients. First year, just $200K."
A few people clapped. Lena didn't. Her title was Senior Test Architect, and she'd held it for two and a half years. She knew Vince's style — brief the execs first, tell engineering later. He wasn't asking for opinions.
She was assigned the technical evaluation. Three weeks.
Week one, she went through DeepCover AI's docs. The AI engine sounded impressive. The white paper was even more polished than the product.
Week two, she loaded real production data (anonymized) and ran it through.
Week three, she confirmed three things:
First — DeepCover AI's "AI test generation" had no real reasoning model under the hood. It was a hardcoded rule template engine with a lightweight LLM wrapper. Give it a login page, and it'd template-match "correct password," "wrong password," "empty password." Give it a cross-border payment flow, and it couldn't identify a SWIFT code or a settlement cycle, let alone generate test cases for "rate-fluctuation retry limits." The template library didn't have that branch written, and the LLM couldn't fill the gap.
Second — DeepCover AI was pure SaaS, shared-tenant architecture. NexPay's financial transaction data would route through AWS Oregon. NexPay's compliance policy stated in black and white: financial data must not leave the country.
Third — and this was the killer. The $200K quote was priced "per-environment." DeepCover AI counted each deployment environment as a separate instance. Standard delivery required 3 environments (dev / staging / prod). First-year actual cost: $600K+, with hidden data storage surcharges kicking in from year two.
Lena wrote it up. Four pages. Data, screenshots, comparison tables. Clean.
Vince read two pages and looked up at her:
"Take another look. Don't play it too safe."
Lena took the report back. Walked to her desk. Changed nothing.
She switched strategies instead.
Knife One
Monday. Everett — Director of Operations, the one who handled the company budget — stepped out of a cross-functional sync and ran into Lena in the hallway.
She didn't hand him the report. She just said: "Hey, I went through the DeepCover AI pricing. The structure doesn't quite match our budget template. Might want to get Finance to take a look ahead of time. I'd hate for it to come up in the quarterly review."
That afternoon, Everett pulled up the quote. The structure was indeed off. He forwarded it to the CFO.
The next morning, the CFO — the man who held the company's purse strings — didn't come looking for anyone. He had his financial analyst run the numbers: per-environment × 3, hidden storage fees added in, Vince's "$200K" and the real figure side by side on one page.
Thursday's budget meeting. The CFO pushed his glasses up and put the comparison table on the big screen:
"A $600K project — and Vince said it was $200K. Security flagged it this morning too. Vince, figure out how you want to handle this."
Vince said nothing.
Lena wasn't in that room.
Knife Two
Same week, Wednesday afternoon. The conversation in the break room was never recorded in any meeting minutes.
Lena walked in with her mug. Raj, the security architect — a man who practically slept with the compliance handbook — was already there.
Lena said, offhand: "Hey Raj, has your team looked at DeepCover AI yet? Their docs say it's SaaS deployed, data goes through AWS Oregon."
Raj frowned. "Oregon?"
"Yeah. us-west-2, I think."
Raj didn't reply. He refilled his water and left.
The next morning, security review flagged a red light. Raj's email went to CTO, CFO, VP of Engineering. Three lines:
"DeepCover AI deployment: AWS Oregon (us-west-2). NexPay compliance policy section 4.3: financial data must not leave the country. Unless the vendor provides an on-premise deployment within jurisdiction, security review cannot pass."
Vince replied with two words: "Got it. On it."
Vince knew Raj's email had landed in the CFO's inbox two hours before the budget meeting.
Knife Three
Vince had scheduled the final Demo before Lena even submitted her report. Now he wanted to use it to push past the CFO's and Security's concerns in one clean stroke.
He'd set the Demo for two weeks after the report. DeepCover AI's sales team flew in with their best engineer. Vince wanted to prove to everyone in the room that this was the right call.
Lena was assigned to support the Demo environment setup. Routine work — the day before, she'd filed a standard regression test data update in JIRA, adding the Q2 multi-currency transaction scenarios that had just gone live. Ticket filed, PR submitted, code reviewed. Nothing hidden.
Demo day. The conference room was full — Vince, Everett, Raj, the CFO, and two board observers.
DeepCover AI's engineer ran the standard Demo first. Three happy paths, full coverage, clean. Vince's shoulders relaxed.
The standard run finished. The engineer started closing his laptop.
Lena looked at the screen, almost to herself: "Q2 multi-currency went live last week. Not in their test suite yet."
Everett picked it up. "Hold on — run it against real transactions."
The engineer paused. "We can try that."
Vince nodded. The engineer pulled the latest regression data from NexPay's staging environment.
First batch — 92% coverage. Still fine.
Second batch — 87%. The engineer frowned and started tweaking parameters.
Lena didn't say a word.
Third batch — 74%. DeepCover's template library had no match for "multi-hop exchange rate + partial refund + cross-day timezone." The lightweight LLM started hallucinating. A result flashed on screen: a $12,470 real transaction, calculated as $743.60.
The CFO closed his laptop.
But it wasn't over. One of DeepCover's generated test cases triggered an infinite loop — a settlement failure resubmitted over and over. Three minutes, 4,000+ API calls. The staging environment crashed. Slack alerts exploded. Monitoring dashboards turned red.
Vince stood up. "Restore the environment first."
No one was looking at the screen anymore.
She'd known the outcome three weeks ago. Vince didn't know. DeepCover's engineer didn't know. They were just using "the latest data."
Lena walked back to her desk. She hadn't done anything — she'd filed a routine regression test update. DeepCover AI had lunged at the bait and choked on it.
No One Cared Where the Knives Came From
The DeepCover AI deal was dead. The CFO didn't approve the budget. Security didn't lift the red flag. Screenshots of the Demo crash were circulating on Slack.
Vince never brought it up again. Not because he didn't want to know — because knowing would have been worse.
Everett's stock went up half a notch with the CFO, and his relationship with Security tightened too. He bought Lena a coffee. "Good work."
Lena wasn't sure what he meant. She just nodded.
She sat on the train home.
She thought about Vince's words — "Don't play it too safe." He wasn't teaching her about testing. He was teaching her about power. He just didn't know it.
She thought about DeepCover AI's template engine, those test cases that couldn't recognize a SWIFT code. This was the best AI testing platform on the market — what if someone actually built the right solution?
Three weeks ago, she'd written a report. Four pages, every word correct. Vince sent it back.
Three weeks later, she'd written nothing. Three knives came from three directions, and the people holding them had no idea about each other. Not one knife pointed back at her.
A thought surfaced: What if I'm not just stopping bad purchases — what if I get to pick my opponents?
She opened her notebook. Not a technical plan. A company name. She hadn't told anyone yet.
This is what "Kill with a Borrowed Knife" means — when you can't make the move yourself, let someone else make it for you.
🤖 AI Post-Mortem
[36 Stratagems Tactical Database v3.1] Loaded
[Tactic Match] Kill with a Borrowed Knife
[Analysis Mode] Full-field scan
━━━━━━━━━━━━━━━━━━━━
Tactic match: 91.5%
Operator: Lena
Action: No direct opposition to CTO decision. Used finance, compliance, and the system itself — three independent forces — to block a flawed AI platform purchase
Objective: Stop DeepCover AI procurement
Result: Achieved — deal terminated, CTO never traced the source
Three-knife analysis:
- Finance: A hallway remark → CFO discovers 3x actual cost → budget veto
- Compliance: A break room remark → Security finds data leaving jurisdiction → compliance red light
- System: Routine regression data update → Everett calls for real-data test → template engine can't match → coverage 92%→74% → LLM hallucination + infinite loop → Demo crash
Verdict: Textbook execution of "Kill with a Borrowed Knife." Three knives from three directions, the holders unaware of each other. No blade points back at the one who borrowed them.
Next stratagem: Wait at Leisure While the Enemy Labors
P.S. English isn't my first language. I use AI to polish the writing and smooth out the rough edges. Thanks for reading. ☕ Buy me a coffee
Top comments (16)
Another good read 😄
I like how the stories don't feel repetitive even though they're part of the same series. Curious to see what you do with the next stratagem.
Appreciate it 🙏 That's actually the hardest part of writing this series — each protagonist thinks differently, so the story has to feel different too. The next one's been a rewrite loop, but I think it's finally clicking. Hope it lives up to the wait 😄
Loved the storytelling! 😄 The "borrowed knives" metaphor made the whole AI negotiation feel like a crossover between a startup pitch and an ancient strategy game.
My favorite part is how the story shows that in AI, the smartest move isn't always building the biggest model—sometimes it's borrowing the right strengths at the right moment.
Also, Lena deserves an award for surviving what felt like an AI version of Game of Thrones... except with APIs instead of dragons.
Great read—looking forward to the next stratagem!
Right? Fellow GoT fan here 🙌 (PS: last season doesn't count 🫣 — I still haven't forgiven the showrunners for that one). When you brought it up, the theme song instantly started playing in my head 🎵 Lena's definitely the "win by any means" type — clean, dirty, she doesn't discriminate. As for what's next... The stratagem is coming.
Haha, I completely agree. 😄 The moment someone mentions GoT, that iconic theme starts playing in my head too.
And yes... I think Lena is exactly the kind of character who doesn't care whether the path is clean or messy—as long as it gets the job done. That unpredictability is what makes her interesting.
Now you've definitely made me curious about the next stratagem. Looking forward to seeing how the story unfolds!
About Lena — I was gonna pick up right where her last story ended, but I think seeing where she came from makes her hit harder. So I've been rewriting the next one. Once the first six solo pieces are out, I'll put together a little prelude to tie them all together. Stay tuned 😝
The detail that makes this more than a story: DeepCover passed the demo because the demo was the surface. Three happy paths, clean, a whitepaper glossier than the product. The operating layer a template engine with an LLM wrapper that can’t read a SWIFT code only showed up when real data hit it. $12,470 read as $743.60 isn’t a bug, it’s the surface and the system finally disagreeing in public. On your own question, manipulation or leadership I think you put your finger on the line yourself. Lena never changed the facts. She changed which layer people were looking at. Her four-page report was correct and it died, because a correct report is a receipt, and Vince could send a receipt back. The crash was an outcome, and you can’t send an outcome back. She didn’t win by lying. She won by moving the argument to the layer where the truth was already standing. That’s not manipulation. That’s refusing to argue on the surface someone else picked for you.
This is exactly the dynamic the series tries to get at — the difference between knowing the truth and being able to prove it in someone else's language. Lena could have told the board about the SWIFT parser forty times. Each time Vince would have had a reply ready, because he controlled the vocabulary of the conversation. What she needed wasn't a better argument. She needed the system to produce evidence in a language Vince couldn't talk his way out of: downtime. A crashed production environment is the one document that doesn't need a counter-signature.
"Refusing to argue on the surface someone else picked for you" — that's going in my notes.
“Doesn’t need a counter-signature” that’s it exactly, and it’s a sharper way to say what I was circling. A report is an argument, and arguments live in whoever controls the room’s vocabulary. A crash isn’t an argument. It doesn’t ask the CFO to agree with anything, it just happened, in front of everyone, at the same time.That’s maybe the real definition of an outcome versus a receipt: a receipt still needs someone to interpret it favorably. An outcome doesn’t ask permission. Lena didn’t out-argue Vince she just stopped competing on a field where he owned the ref.
You're right and I think that's exactly the borrowing at work. Lena didn't bring a weapon to the fight — she let the system hand her one. The crash wasn't caused by anything she did. It was always going to happen. She just had her receipt ready when it did. That's the borrowed knife: an outcome that was already on its way, arriving while she was standing there with the only document that still made sense.
The ref-owned-field line hits something deeper that I've been trying to name across all six of these. Every protagonist walks into a room where someone else wrote the rulebook — Mark, Derek, Lena, the next three. The interesting thing is they don't all take the same way out. Mark outplayed the rules. Derek let the rules collapse on themselves. Lena stepped off the board entirely. The series isn't about one winning strategy. It's about reading which game you're in, then deciding what "winning" even means from that seat.
I’ve read the first four in the series, and this is my favorite so far.🔥
Looking forward to the next ones!🙌
haha right? Lena's my favorite too🤣 — her depth is on a completely different level from the other protagonists. You'll see more of her🙌
What strikes me is that Lena never lies. She just... rearranges what people see. Is that manipulation or leadership? Genuinely not sure where the line is.
Another banger, once I start reading these stories I can't stop, need to finish them!
Appreciate that man 🙌 That's the best compliment you can give a storyteller — when someone can't stop. Got a few more in the pipeline. 👀
Halfway through the first 6 — 6 protagonists, 6 solo stories, zero crossovers. 3 down, 3 to go. Keeping the keyboard smoking 🔥