Intent-Driven Engineering: What Happens After the Hype
- Mark Kendall
- 1 minute ago
- 2 min read
Title
Intent-Driven Engineering: What Happens After the Hype
Intro
Everyone is talking about how fast you can build with AI.
Fewer people are talking about what happens when you actually try to build something real.
I just spent days pushing an AI system to generate a working spatial application—maps, geocoding, narrative layers, the whole thing. It worked. Until it didn’t.
And that’s where the real lesson starts.
What Is Intent-Driven Engineering?
Intent-Driven Engineering is the practice of describing what you want a system to do in natural language, and letting AI generate the implementation.
It feels like magic at first.
You describe a feature, and the system builds it.
You describe a bug, and the system fixes it.
But there’s a catch.
The Reality: You Can Build Fast… and Break Faster
In my case:
I had a working map
Points were rendering
Geocoding was accurate
Then I asked for something simple:
“Minimize the sidebar so I can see more of the map”
That’s it.
After that:
Points disappeared
Geocoding failed
API errors started showing up
Nothing else changed.
This is the moment most people hit and don’t talk about.
The Core Lesson: Intent Must Be Iterative
At first, I went for everything:
Full feature set
Complete experience
“Build me the system”
And the AI tried to deliver.
But every step:
Consumed tokens
Introduced complexity
Increased fragility
What I learned:
Intent is not a one-shot command. It’s an iterative discipline.
Why Iteration Matters (More Than Ever)
In traditional development:
You iterate to manage complexity
In AI-driven development:
You iterate to manage cost, stability, and control
Without iteration:
You overbuild
You overspend
You lose track of what broke
The Guardrails I Learned the Hard Way
1. Always Maintain a “Known Good State”
If something breaks:
Don’t fix forward
Roll back
Stability is more valuable than progress.
2. One Intent at a Time
Bad:
“Build the full system with all features”
Good:
“Fix geocoding”
“Render points”
“Add sidebar toggle”
3. Observe Everything
Use logs:
API responses
Coordinates
Rendering state
If you can’t see it, you can’t control it.
4. Separate Concerns Early
Don’t try to do everything in one layer.
Example:
UI = interaction
Export = separate system
Trying to combine them leads to chaos.
5. Control Your Spend
AI will try to do everything you ask.
It doesn’t:
manage your budget
optimize your tokens
protect you from overbuilding
That’s your job.
Why Most People Haven’t Hit This Yet
Right now, most content is still in:
“Look what this can do”
Not:
“Here’s what happens when it breaks”
But if you’re building real systems, you’ll hit this stage quickly.
And when you do, you’ll realize:
The power isn’t in generating code.
It’s in controlling how it evolves.
Why This Matters
This isn’t about maps.
This is about a shift in how software is built.
We’re moving from:
writing code
to:
directing systems
But direction requires discipline.
Key Takeaways
Intent is not magic—it’s a process
Iteration is mandatory, not optional
Stability beats speed
Visibility beats assumptions
Control beats capability
Final Thought
AI didn’t fail.
My expectations did.
Once I adjusted how I approached intent, everything changed.
Not slower.
Not harder.
Just more controlled.