We Stopped Writing Code and Started Engineering Context
Two years ago, our engineering teams were drowning.
We had powerful AI models… but we used them like fancy autocomplete. We celebrated 3× velocity but quietly ignored the mounting “AI slop.” We shipped faster but understood less. We weren’t engineering, we were just asking AI to guess.
The breakthrough wasn’t GPT-6, Claude-Next, Cursor 2.0+ or any new model.
The breakthrough was Context Engineering the discipline that turned AI from a chaos generator into a force multiplier.
1. The Evolution of Code Review:
From "Check Style" → to Align Minds
In early AI coding, our goal was simple: ship faster.
But Dex Horthy nailed the truth at the AI Engineer Code Summit — we did ship 2–3× more code… and 2–3× more rework. Because we were reviewing the wrong thing.
Style? Automated. Bug-finding? Automated. Mental Alignment? That became everything.
By 2026, I don’t read every line of generated code. I read the plan.
As Dex said:
“I can read the plans… I can catch problems early… I maintain understanding of how the system evolves.”
A bad line of code is cheap. A bad plan is catastrophic. That was the mindset shift.
2. The “Seniority Rift” of 2025
When Mid-Levels Loved AI… and Seniors Wanted to Quit
This was the graph that woke up every CTO.
- Mid-Level Engineers: thrilled — AI filled skill gaps and boosted velocity.
- Senior Engineers: miserable — forced into janitor roles cleaning up endless AI-generated chaos.
They didn’t hate AI. They hated becoming sweepers of AI slop. The core issue?
We were outsourcing thinking. Not tasks.
No AI tool can understand a 10-year-old brownfield codebase for you. But once you understand it, AI can amplify you 100×.
This cultural rift forced us to define a new rule:
AI is not a brain replacement. It’s a cognition amplifier.
Once teams understood that, the rift closed.
3. The Architecture of the Agentic Age
Prompt Engineering Was Just 5% of the Real Job
Recommended by LinkedIn
Context Engineering
We learned:
- The Context Window is a finite resource.
- At ~40% usage, you enter the Dumb Zone where models hallucinate more.
- To stay in the Smart Zone, you use Intentional Compaction — compressing the truth of the system.
That’s where the RPI loop emerged:
🔍 Research
Map the system. Extract the truth. Compress it without losing meaning.
📝 Plan
Define steps, constraints, validations, exact snippets.
🛠️ Implement
Let the model generate code — now failure-resistant because context is curated.
This loop became the default workflow for every serious AI Engineer.
We stopped “chatting with AI.” We engineered the environment where AI can’t get it wrong.
4. The Leadership Mandate
You Cannot Grassroots a Revolution of This Scale
This was the hard truth:
Cultural change requires top-level buy-in. Full stop.
The teams that succeeded in 2025 weren’t the ones with the flashiest tools. They were the ones where Staff+ engineers actually used AI themselves.
Not to write less code… but to understand where the foot-guns were.
They practiced. They failed. They got better. And then they taught the rest of us.
That leadership changed everything.
Today, 99% of our code is AI-generated. But 0% of our architecture is.
Humans design intent. AI executes it. We’re not doing vibes-based development anymore.
We’re doing Context Engineering — and it brought back craftsmanship in a world drowning in automation.
If you’re still treating AI agents like chatbots…
You’re not behind. You’re living in the wrong era.
It’s time to engineer the context — because the future belongs to the teams who think clearly, not the teams who type fast.
Reference: Dex Horthy – No Vibes Allowed: Solving Hard Problems in Complex Codebases AI Engineer Summit https://www.youtube.com/watch?v=rmvDxxNubIg