Everyone’s excited about vibe coding. Nobody wants to talk about systems thinking.
I’ve been having more conversations about AI than ever. They keep ending the same way.
Someone brings up ChatGPT. Someone else counters with Gemini. A third person mentions they vibe-coded an app over the weekend. Everyone has opinions. Nobody can explain what’s actually happening beneath the interface they’re using.
This isn’t a knowledge gap. It’s a discourse gap. People aren’t debating capabilities or architecture or tradeoffs. They’re comparing brand preferences and calling it expertise. And in Philippine tech circles, where the energy around AI is real and the ambition is genuine, this confusion is doing actual damage.
01The Guru Promise vs. the Invoice
Social media is full of promises right now. Every other post is about how AI has been democratized, how anyone can build anything, how the tools are free and the future belongs to whoever moves fastest.
Here’s what those posts leave out.
“Free and open source AI models!”
Moonshot AI released Kimi K2, an open-weight model with a trillion parameters. Free to download. The catch: running even the most aggressively quantized version locally requires a minimum of 247GB of combined memory. The full-quality version? Over 360GB. NVIDIA’s Nemotron 3 Ultra, another “open source” release, packs 550 billion parameters and requires multiple enterprise-grade GPUs to run inference. Most developer laptops ship with 16GB of RAM.
Free means the weights cost nothing. Running them costs more than most people’s entire hardware setup.
“Build an app without coding!”
Replit and Lovable are the tools of choice in PH dev circles right now. They’re good at what they do: spinning up a working prototype fast. But a prototype is not a product. Shipping that prototype means you need a server, a domain, SSL certificates, a database, error handling, and a plan for what happens when the template breaks at 2 AM. That’s not a one-click operation. That’s infrastructure, and it comes with recurring costs and maintenance you won’t learn from a YouTube tutorial.
“AI replaces developers!”
AI generates code. It doesn’t debug code when the generated output conflicts with your existing system. It doesn’t know your deployment target, your security requirements, or which dependency just pushed a breaking change. The code it writes needs a human who understands the system it’s landing in. That human is still a developer. Now they’re also debugging AI output on top of everything else.
“Just ask ChatGPT. It knows everything!”
I ran a frontier AI model on a structured SEO audit this week. Not a casual prompt. A specific, scoped task with clear inputs. It returned four findings. Three were fabricated. It reported a file had 16 entries when it had 18. It claimed a search ranking of #3 for a result that wasn’t in the top 10. It cited a company in the SERP data that simply wasn’t there. One finding out of four held up.
This wasn’t a free-tier chatbot. This was a frontier model doing scoped, professional work. It still hallucinated through 75% of its output. If you’re not re-reading and cross-checking every response, you’re absorbing a mix of fact and fiction and calling it research.
02What Vibe Coders Skip: Systems Thinking
The gap isn’t between “knows AI” and “doesn’t know AI.” It’s between “can use a tool” and “understands how things connect.”
Systems thinking is the practice of seeing dependencies, failure modes, and consequences before they hit production. It’s not an academic framework. It’s the difference between building something that works in a demo and building something that survives contact with real users.
| Vibe Coding Assumption | Systems Thinking Reality |
|---|---|
| “My app uses an API” | Your app depends on a third-party service with rate limits, pricing changes, and a TOS that can shift tomorrow |
| “It works on my machine” | What happens when the API is down, the token limit hits, or the model returns garbage? |
| “The data shows up” | Where does this data live? Who owns it? What happens when the browser tab closes? |
| “I deployed it” | The API key is in the frontend. Anyone with browser dev tools can read it. |
| “It’s free to use” | Free tier has limits. Production usage has a bill. That bill scales with every user you add. |
I wrote about what happens when vibe-coded projects hit production without these foundations. The pattern keeps repeating. Prototype works. Demo impresses. Production breaks. And the person who built it can’t diagnose why, because they never learned how the pieces connect.
03Before You Ship: Five Questions Your Vibe-Coded Project Should Answer
You don’t need a computer science degree. You need to answer these five questions about your own project before you put it in front of anyone:
- What happens when your main dependency goes down? If you can’t name your main dependency, start there.
- Where does your data live, and who can access it? “In the cloud” is not an answer.
- What does your app cost to run per month at 100 users? At 1,000? If you don’t know, you haven’t priced your own product.
- Can you read the error log and find the problem? If the AI wrote the code and you can’t trace the failure, you’re stuck the moment something breaks.
- What changes if your AI provider updates pricing or rate limits tomorrow? If the answer is “everything breaks,” you’ve built on sand.
These aren’t trick questions. They’re the minimum bar for shipping something real. If you can answer all five, you’re already thinking in systems, whether you call it that or not.
04Frequently Asked Questions
Is vibe coding bad?
No. Vibe coding is a legitimate way to prototype and explore ideas. The problem starts when people treat a prototype as a finished product, or when the person building it can’t explain how it works. The gap between “it runs in Replit” and “it runs in production” is exactly where systems thinking lives. Use vibe coding to explore. Use systems thinking to ship.
Do I need to learn traditional programming to use AI tools effectively?
You don’t need to write code from scratch. But you need to read code, trace errors, and understand how the pieces of your system connect. AI tools generate output. Understanding that output, knowing when it’s wrong, and fixing it when it breaks requires foundational literacy that no browser-based AI hands you automatically.
What is systems thinking in software development?
It means understanding how parts of a system interact, depend on each other, and fail. In practice, that includes knowing your dependencies, your failure modes, where your data lives, and what your security surface looks like. It’s the difference between building something that demos well and building something that holds up when real people use it.
Can AI output be trusted for professional decisions?
AI models are tools, not authorities. Even frontier models with safety guardrails fabricate data during structured, professional tasks. Every AI output should be verified against primary sources before you act on it. The higher the stakes, the more critical the verification step becomes.
05Vibe Coding Without Systems Thinking Is a Dead End
The tools are real. The energy in Philippine tech is real. But the shortcut from “I use ChatGPT” to “I’m an AI expert” skips the part that actually matters: understanding the systems you’re building on.
If you’re evaluating someone who claims AI expertise, ask them about failure modes, not features. Building with AI yourself? Run through the five questions above before you ship. The gap between tool fluency and systems literacy is where projects go to die.
That gap doesn’t close with better prompts. It closes with better infrastructure thinking.
Tom Tokita is President of Aether Global Technology Inc., a Salesforce consulting firm in the Philippines. tokita.online is where he documents his personal AI R&D: building and maintaining production AI systems that route across multiple LLMs daily.