One thing I want to make clear upfront:
I do not hate AI.
Honestly, I think AI is one of the most important technological shifts we’ve seen in decades.
I use it.
I study it.
I think it will dramatically change:
- software development
- infrastructure
- operations
- education
- research
- business workflows
But at the same time, I also think the current culture surrounding AI is becoming dangerously misguided.
Not because the technology itself is bad.
Because the mentality around it increasingly feels shortsighted.
⸻
The Current Conversation Often Feels Backwards
A lot of AI discussions right now revolve around:
- replacing engineers
- generating apps instantly
- removing expertise
- eliminating learning curves
- “vibe coding”
- shipping software without understanding systems
And honestly?
I think that framing is incredibly dangerous long-term.
Because software is not just text generation.
Software becomes:
- infrastructure
- operational systems
- financial systems
- healthcare systems
- logistics systems
- communication systems
- ecosystems people depend on daily
The consequences of poorly understood systems become very real very quickly.
⸻
The “Vibe Coding” Mentality Worries Me
One thing that concerns me is how casually some people now approach building operational systems with AI.
There’s this growing idea that:
“If the application appears to work, understanding no longer matters.”
But operational systems are not just:
- UI screens
- generated routes
- copied prompts
Real systems involve:
- scalability
- security
- lifecycle management
- deployment
- observability
- infrastructure
- maintainability
- operational boundaries
- data integrity
- runtime behavior
Those things still matter enormously.
And honestly, they may matter more now.
⸻
Earlier Prompts Already Feel Like Legacy Code
One thing I find fascinating is how quickly AI-generated workflows are already aging.
Prompts from:
- earlier models
- earlier workflows
- earlier tooling patterns
already feel like operational legacy systems.
That should probably tell us something.
Because if the workflow itself changes every few months, then:
- architecture matters
- operational clarity matters
- maintainability matters
- foundational knowledge matters
Otherwise systems become extremely fragile extremely quickly.
⸻
Token Cost vs Engineering Quality Is a Real Conversation
Another thing I think people underestimate is the operational cost side of AI.
There’s a growing assumption that:
“AI is automatically cheaper than engineers.”
But in many cases:
- debugging generated systems
- correcting architectural mistakes
- dealing with scaling failures
- untangling hidden complexity
- fixing poor runtime decisions
can become incredibly expensive operationally.
Especially when systems move beyond prototypes.
Sometimes paying experienced engineers to:
- design systems properly
- establish operational boundaries
- create maintainable architecture
- think through lifecycle implications
is dramatically cheaper long-term than repeatedly generating unstable systems quickly.
That’s not anti-AI.
That’s operational realism.
⸻
Tools Still Require Stewardship
One analogy I keep coming back to is this:
Just because someone has access to a powerful tool does not automatically mean they understand how to wield it responsibly.
And honestly, I think AI is one of the most powerful tools we’ve ever placed into people’s hands.
That means:
- training matters
- understanding matters
- responsibility matters
- operational thinking matters
Especially when these systems increasingly affect:
- businesses
- infrastructure
- communication
- financial systems
- public systems
This is bigger than generating websites quickly.
⸻
AI Should Amplify Engineers — Not Replace Understanding
Personally, I think the healthiest future is one where AI amplifies:
- engineers
- architects
- operators
- creators
- educators
instead of convincing people foundational understanding no longer matters.
Because understanding systems deeply still matters.
Maybe now more than ever.
Especially as generated complexity increases.
⸻
The Real Bottleneck Was Never Typing Speed
One thing I think the industry is slowly realizing is that software engineering was never primarily bottlenecked by:
- typing speed
- boilerplate generation
- syntax production
The harder problems are usually:
- architecture
- operational clarity
- responsibility boundaries
- scalability
- maintainability
- communication
- infrastructure
- lifecycle sustainability
AI helps with implementation.
But implementation was never the entire problem.
⸻
This Is Why I Keep Thinking About Explicit Systems
A lot of my current thinking around:
- WebEngine
- KiwiPress
- operational architecture
- contracts
- pipelines
- blueprint systems
comes from this exact concern.
The more powerful generation becomes, the more important:
- clarity
- structure
- observability
- maintainability
- operational boundaries
become.
Otherwise we risk generating operational chaos at unprecedented speed.
⸻
I Think We Need Better AI Culture
I don’t think the answer is fear.
And I don’t think the answer is rejecting AI.
I think the answer is developing a healthier engineering culture around it.
One that values:
- education
- stewardship
- architecture
- operational understanding
- maintainability
- responsible system design
instead of:
- hype
- shortcuts
- replacing understanding
- generating systems blindly
Because these tools are becoming too powerful to treat casually.
⸻
Final Thoughts
AI is not the problem.
The mindset surrounding it often is.
I think AI has the potential to become one of the greatest engineering accelerators we’ve ever seen.
But acceleration without understanding can become dangerous very quickly.
Especially when software increasingly powers:
- businesses
- infrastructure
- operations
- communication
- daily life
The future shouldn’t be:
“nobody needs to understand systems anymore.”
If anything, I think the future requires deeper operational understanding than ever before.
Because the more powerful our tools become…
…the more responsibility comes with using them well.
Top comments (0)
Some comments may only be visible to logged-in visitors. Sign in to view all comments.