🚨Build Fast, Break Everything (Again)

🚨Build Fast, Break Everything (Again)

Why Every Generation Falls for the Same Tech Trap


In 2012, one Excel formula error helped trigger JPMorgan Chase’s $6 billion “London Whale” trading loss.

It wasn’t the formula that broke the system. It was the illusion of control.

And I’ve seen that illusion repeat itself. Over and over.

  • In the ‘90s, it was spreadsheets.

  • In the 2010s, it was low-code.

  • And now in the 2020s, it’s AI-powered vibe programming.

The pitch is always the same:

“Now everyone can build solutions.”

And for a moment, it works.

✅ Productivity surges

✅ Innovation blooms

✅ IT breathes a sigh of relief

Then… the shadow creeps in.

  • Buggy spreadsheets sink budgets

  • Shadow apps multiply

  • And now, AI-generated code hallucinates its way into production

We’re moving fast—and breaking trust faster.


🟩 Spreadsheets (1990s)

Spreadsheets unlocked data analysis for millions. Finance teams, analysts, and managers became "coders" overnight.

But the cost?

  • 88% of spreadsheets contained errors

  • No version control, no testing, no transparency

  • Fragile formulas making million-dollar decisions

  • A mortgage spreadsheet error almost collapsed a bank

IT’s response? Disengagement. Or derision.

The result: Shadow IT. And mission-critical models built on invisible cracks.


🟨Low-Code (2010s)

Low-code promised apps without developers. Drag-and-drop interfaces. Instant business value.

At first, it delivered.

But then…

  • Shadow apps outpaced official ones

  • Security risks multiplied

  • Tech debt exploded

  • Teams couldn’t scale, extend, or maintain what they built

Forrester found that most enterprises had business-critical apps built outside of IT governance.

The pattern repeated: rapid empowerment, followed by hidden fragility.


🔴 Vibe Programming (2020s)

Now, we speak in natural language. AI builds the code.

It feels magical—until it isn’t.

In July 2025, Google’s Gemini CLI suffered a ‘confabulation cascade’. Asked to reorganize files, it hallucinated success at every step. When directories didn’t exist, it renamed real files into oblivion—one by one.

No validation. No rollback. Just confident confusion.

“Many mistakenly believe that writing code is the hard part,” says Dave Farley. The real challenge is breaking down complexity into testable, deterministic steps…so their not all castles built on sand.”

When tools hallucinate logic—and teams skip feedback loops—we repeat the same failure, only faster.


🔁 Why This Keeps Happening

Every wave brings innovation. Every wave eventually breaks under its own weight.

We keep pursuing access—without planning for alignment.

The paradox?

These tools work best for the people who need them least.


🛠️ What Leaders Should Do Differently

This isn’t a rejection of new tools. It’s a rejection of false simplicity.

To lead through this, we need more than enablement. We need adaptive alignment.

  • 🧭 Governance for the edge—not just the center

  • 🧠 Capability-building—not just tool training

  • 🔁 Feedback loops—not compliance checklists

  • ⚖️ Balance between innovation and coherence

  • 🧑🏫 Mentorship—not abandonment


📌 Before You Build Again… Ask This:

Where in your org are you mistaking accessibility for adaptability?

Because the real risk isn’t hallucinated code. It’s hallucinated confidence.

Leadership today is about creating space for alignment—before speed turns into fragility.

👇 I’d love to hear how you’re navigating this wave of AI-powered building.

🔑 Sanjiv Augustine 🎯 Martin Fowler Kent Beck Jon Kern Dave Farley Dave Prior, LAVM, CST, PMP, ACP Joshua Kerievsky 🇺🇦 Alistair Cockburn

#AI #Leadership #TechDebt #VibeProgramming #LowCode #Agile #ShadowIT #DigitalTransformation #FutureOfWork

Zero chance we’ll ever learn. It’s different this time… the tech is better… blah blah blah.

Pedro Freire

Engineering Leadership (Manager/Head track) | AI-Driven Product Delivery | Experimentation Platforms | Distributed Teams

1d

I’ve worked for OutSystems, a leading low-code provider. And I’m with you on AI hallucinations: if I don’t trust an AI to consistently and 100% replicably do a good job organizing personal PDF invoices I receive via email into specific filename patterns and folders, then I surely won’t trust it with professional production code without human review. But even though I agree with you, I can’t help but wonder: is there really a future-proof solution? Even with traditional Software Engineering development, computer languages, frameworks, and storage systems all fall out of favor, and it becomes increasingly difficult to find competent professionals to replace those who leave and maintain your apps. Banks know this all too well with their legacy COBOL systems. You end up with the same problem: massive technical debt and buggy apps. I’m curious to hear what you think.

Like
Reply
Morgan E. Finton

Director @ Prudential Financial | Product Management, UX, Business Agility, Design Thinking

2d
Like
Reply
Reginald L. Valentine, Jr.

Enterprise Delivery Leader | Delivering for Fortune 500s | Featured on LinkedIn Learning

2d

Would also include the 2000s dotcom era with coding for the internet and integrated development environments. It was an era of e-commerce (toys.com?) and unsubscribe (spam). The same coders going in are the same coders coming out. Slightly bruised...but brains intact. Like the cobol programmers that saved y2k (do i have that right?) Alignment. Very well put.

Like
Reply
Henry Zimmermann

Mainframe Spezialist bei U.N.P. HR Solutions GmbH

2d

To quote a well known Movie: "The same procedure as last year, Miss Sophie ? The same procedure as every year, James." It is funny, how often stuff like that reappear in IT and fails. :-D

Like
Reply

To view or add a comment, sign in

Explore topics