I had the privilege to attend this mini-conference on AI and Software Engineering. It absolutely changed my mind regarding the role of AI and that of software engineers. I walked into the room expecting the usual industry talking points.
I expected another round of defensive arguments about why developers won't be replaced, or surface-level demonstrations of code generation.
Instead, I got an absolute paradigm shift.
Listening to industry veterans like Amariah Abishai (Founder, Atlarix) and Paul Oguda (Lead Technical team, Zone01Kisumu),
watching our top apprentices like @bwanachairman , @gnyaboke , Mercy Moraa, and Joy Lamka break down the realities of building in production, and diving into workshops with Michael Clay and Richard Ochola triggered a profound realization.
The endless internet debate over whether AI is a "superpower or a crutch" is completely dead.
While we were busy arguing about the tools, the entire software engineering stack quietly moved up.
We didn't stop being developers. We all just got promoted to architects.
The true challenge facing modern engineers is no longer about learning how to type code faster.
It is navigating the razor-thin, high-stakes line between Vibe Coding and Viable Coding.
*What is "Vibe Coding"?
*
Coined loosely in the AI research spaces, Vibe Coding is a state where a developer inputs high-level natural language prompts, watches an AI agent generate a mountain of files, and watches an application spring to life.
At that particular moment, it feels magical.
It is fast, highly creative, and immediate.
For rapid prototyping, it is an undeniable superpower.
But left unchecked, it becomes a dangerous crutch.
When you vibe code, you outsource your foundational logic to a probabilistic model. The AI will happily write code that looks beautifully plausible, but plausible and correct are not the same word.
Vibe coding creates a false sense of security.
It gives birth to the "Spaghetti Prompt Effect" —where a developer continuously patches broken AI code with more AI code until the underlying architecture becomes an un-debuggable black box
*What makes it "Viable Coding"?
*
Viable Coding is where the actual software engineer lives.
It is the transition from being a bricklayer to being the architect.
An architect does not measure their value by how many lines of code they manually type per minute.
They measure it by judgment.
The moment a model handles the syntax, your job shifts to asking the hard questions:
- What is actually worth building, and why is it a game-changer?
- How does this system hold together at the seams?
- What should we ruthlessly choose NOT to build to keep the scope clean?
*The Multi-Dollar Proof of the Architect Moat
*
If you think the architect layer is just an abstract corporate buzzword, look at how the biggest AI-era software companies are winning right now.
Take Cursor and Perplexity.
Cursor scaled to a $2B annualized run rate by early 2026.
Here is the catch: they did not train a frontier LLM model from scratch. Their editor is powered by models they don't own—like Anthropic's Claude.
Perplexity built a massive category-defining search engine valued at $22B, orchestrating existing foundation models and web retrieval.
Neither of these companies won because they had a secret, proprietary AI engine. They won because they were superior architects around the engine. Their competitive moat was the workflow, the interface, the data flow, and the system design.
The models they use are available to you and me via an API. The gap isn't the technology; the gap is the architectural vision.
*The Context Paradox: Why We Are Irreplaceable
*
During the panel discussions at Zone01Kisumu, a brilliant developer, Kolaka Joe, dropped a piece of food for thought that perfectly frames the limits of pure AI automation:
_"If AI tools are so smart and actively replacing us engineers, why haven't they come up with a cure for cancer?"
_
It is a profound point. AI possesses massive raw compute power and can process endless variations of code. But compute power without human domain context, critical thinking, and structural intent cannot solve deep, systemic, real-world problems.
The machine can lay a billion bricks flawlessly in seconds. But it has absolutely no idea why a building needs to stand, how it anchors into the specific local soil, or who it is protecting.
That requires a human architect.
*Conclusion: You're an Architect Now. What Are You Going to Build?
*
The stack didn't shrink; it just grew a new floor. The barrier to entry to write code has dropped to zero, which means the premium on structural judgment, systems design, and critical engineering thinking has skyrocketed.
We need to stop fighting the tools and start owning the blueprints. Use the AI to move fast, explore, and eliminate boilerplate syntax. But never surrender the architecture.
Keep your eye strictly on the line where the "vibe" ends and engineering viability begins.
The tools are ready. The blueprint is in your hands.
So, what are you going to build?
Top comments (0)