DEV Community

Cover image for Stratagems #5: Leo Walked Into an AI-Powered Burning House. He Walked Out With a Client.

Stratagems #5: Leo Walked Into an AI-Powered Burning House. He Walked Out With a Client.

xulingfeng on July 02, 2026

When the enemy is in distress, exploit the opportunity to seize advantage. — The 36 Stratagems, Loot a Burning House Who's Leo — In the last...
Collapse
 
hemapriya_kanagala profile image
Hemapriya Kanagala

Really enjoyed this one 😄

I've been following the series, and it's been fun seeing how each story explores a different idea. I especially liked the takeaway at the end about preparation meeting opportunity. Looking forward to the next one.

Collapse
 
xulingfeng profile image
xulingfeng

Thanks! That "preparation meets opportunity" line is my favorite part of this one too — Leo didn't plan the outage, but he'd already paid for that fix years ago. The next one's a different flavor. Hope it lands 😄

Collapse
 
sam_tech_e3c30d03221da839 profile image
Sam Tech

Hi Guys, I'm SamTech a Developer Talent Connector. We're currently recruiting experienced Web Developers, Frontend Developers, Backend Developers, Full-Stack Developers, and App Developers for long-term collaborations with an American development team. This is a remote opportunity involving real international projects, with payments available in USD or local currency. If you're interested in learning more, reply to this comment and I'll be happy to share the details.

Collapse
 
xulingfeng profile image
xulingfeng

Wrong place for this, Sam😅🤣

Collapse
 
itskondrat profile image
Mykola Kondratiuk

coordination failure gets logged as capacity problem in the postmortem. more often than not the agents fight over context, not the hardware. takes a while to unlearn the hardware-first instinct.

Collapse
 
xulingfeng profile image
xulingfeng

"hardware-first instinct" — if that ain't the default human reflex I don't know what is. CPU at 90%? must be the CPU. never the code that wrote 4,000 logs per second 😂

Collapse
 
itskondrat profile image
Mykola Kondratiuk

lol exactly - first assumption is always the box. then you start peeling and it's the agent in a tight loop wondering what to do next

Collapse
 
technogamerz profile image
𝑻𝒉𝒆 𝑳𝒂𝒛𝒚 𝑮𝒊𝒓𝒍

This was much deeper than the headline suggests. My biggest takeaway is that successful strategy isn't about creating chaos for competitors; it's about recognizing when existing chaos has already changed customer behavior. Leo didn't force the client to switch—he understood that trust had already been broken and responded with competence, clarity, and reliability. That's an important distinction because it keeps the strategy ethical while making it highly practical. The connection between the ancient stratagem and modern AI consulting was also well executed. It shows that although technology changes, human decision-making around trust, timing, and value remains remarkably consistent. Really enjoyed reading this.

Collapse
 
xulingfeng profile image
xulingfeng

That "ethical distinction" point — honestly the part I wrestled with most when writing this. Leo could've easily come off as a vulture circling CodeForge's corpse. But the line between "looting a burning house" and "showing up with a fire extinguisher" is where the story actually lives. Glad it landed.
Been thinking a lot while writing this series — am I just writing the same pattern over and over? Core stays the same, only the skin changes. Then I look back at texts written thousands of years ago and realize they already said it all. Technology moves forward, but the core never really changed. It keeps cycling, just wearing different clothes. And we keep cycling through the same roles, playing different versions of the same play. Seeing through the surface is the only way to actually get it.
This came out a bit scattered😇 — I'll do a proper wrap-up post after stratagem #6 to connect the dots.

Collapse
 
motedb profile image
mote

The topology analysis failure mode here is really well-observed. An AI optimizer treating middleware as "transparent nodes" because their return values don't participate in routing decisions — that's the same class of bug I've seen in dependency analysis tools that prune "unused" imports without checking side effects.

What makes this hard to prevent is that the optimizer is technically correct about the data flow graph. The fraud check middleware doesn't modify the request body. It just injects headers into a downstream adapter. The topology analyzer can't see that relationship because it crosses a boundary the trace doesn't follow.

This feels like a broader problem with AI-driven refactoring tools in general — they optimize what they can measure, and they can't measure things they don't know exist. The reviewing engineer glanced at it and it "looked fine" because the generated config was syntactically valid and logically coherent. It just happened to be wrong in a way that required knowing about an integration point that wasn't represented in the data the AI had access to.

I'm curious — in the real-world version of this (if there is one), was there any kind of canary deployment or staged rollout that could have caught this before it hit the full payment pipeline?

Collapse
 
xulingfeng profile image
xulingfeng

Hey, sorry for the late reply — I just shipped #6, so I'm only catching up now.
You nailed it. The optimizer was technically right — the fraud middleware really doesn't touch the request body. That's what makes this bug so nasty. It doesn't crash. It just silently stops doing something, and you don't find out until a transaction fails and someone manually traces it back and goes "wait, why didn't this get checked?"
As for canary deployment — nah, and here's why. The change looked like a straightforward perf optimization (cut a "redundant" hop). A canary would've confirmed the numbers looked good and payments were still going through. It wouldn't have caught the missing fraud check unless that canary's test data happened to include a flagged transaction. Which it didn't. Because nobody knew the fraud middleware was gone. The topology analyzer didn't flag it, so nobody thought to test it.
What I took away from this: if your deployment pipeline only validates what your monitoring tools can see, then anything outside their field of view is invisible until it breaks. Canaries, staged rollouts, even chaos engineering — they all depend on you knowing what questions to ask. The hardest bugs are the ones you don't know to look for.
Appreciate you reading it that deep. 🙏

Collapse
 
leob profile image
leob

Just loved how heavy this one was on technical detail - coincidentally, I worked at a payments company myself previously, tangentially related to Adyen even, I was pretty surprised to see that name - what a small world it is! :-)

Collapse
 
xulingfeng profile image
xulingfeng

Anyone who survived the payments industry and willingly identifies themselves as a fellow survivor — you're one of us 😂 What you see as technical depth, I see as PTSD from debugging payment pipelines. Drop the name "Adyen" and the ones who know, know.

Writing this piece felt like being back in those late nights writing test cases for payment projects. Now the late nights are for writing instead. Guess some things never change.

Collapse
 
technogamerz profile image
𝑻𝒉𝒆 𝑳𝒂𝒛𝒚 𝑮𝒊𝒓𝒍

There is no post today!! that's not fair.

Collapse
 
xulingfeng profile image
xulingfeng

Haha, so this is what it feels like to get asked "where's the post" 😂 Stratagem #6 has a little ending to wrap up that I'm still polishing (read: either procrastinating or deep in a Nintendo Switch session). Aiming for tomorrow. No promises but close.🤣

Collapse
 
technogamerz profile image
𝑻𝒉𝒆 𝑳𝒂𝒛𝒚 𝑮𝒊𝒓𝒍

You play the games, I'll cheer you on! 🎮

Collapse
 
xulingfeng profile image
xulingfeng

Fresh account, first thing you do on the platform — and this is what you went with. Interesting call.