In the cyberpunk stories I love, the world is split into two: the “Chromed” (enhanced, fast, inhuman) and the “Ganics” (natural, slow, fragile). For everyone who has seen the show “Cyberpunk: Edgerunners”, you know what I mean!

I am an engineer who embraces the “Chrome.” I love the shiny new tools. I use the tooling, the agents and the super-powered IDEs. But lately, I have realised that while I am moving faster than ever, I am fighting a daily battle not to lose my grip on the wheel.

I stand right in the middle, looking at my own hands, wondering if I am still the one driving in the future.

Seduction of the Chrome

When you spin up an agent like Claude Code and command it to “refactor the authentication module,” and watch it analyse the repo, edit five files, and fix its own bugs in a blur of text, it feels like magic. It removes the friction. It even reeliminateshe need to open the files.

We type “Apply” because the agent’s reasoning sounds logical, not because we have verified the implementation is accurate. We feel the gravity of the easy path pulling at our attention span, whispering that we don’t need to understand the underlying logic anymore: We need to know how to direct the agent.

We are surrounded by tools that promise to make us architects, but if we aren’t careful, they will turn us into passengers.

The Vibe Coding Trap

We are seeing the rise of “Vibe Coding” -- accepting code because it feels right.

Recent studies confirm what many have quietly feared. Research from 2025 (including studies by Anthropic and Microsoft) reveals a troubling trend: while AI assistance can boost task speed, it comes with a heavy “cognitive tax.”

  • The Mastery Gap: In controlled groups, developers who relied heavily on AI scored 17% lower on mastery quizzes about the very code they had just “written.” (Source: 2025 Developer Mastery Study)
  • The Security Blindspot: Vibe coding doesn’t just create bugs; it imports vulnerabilities and hallucinates packages. We risk shipping “average,” insecure code that passes a glance but fails a deep security audit.
  • Cognitive Atrophy: Just as GPS eroded our innate sense of direction, offloading logic to an LLM erodes our “algorithmic fitness.”

The standard defence is: “The models will get better. Soon they won’t make mistakes.”

Which is precisely the problem.

In aviation, this is known as the Irony of Automation. As autopilots became more reliable, pilots became less capable of handling emergencies. When the system works 99% of the time, your brain checks out.

When GPT-5 makes a syntax error, you catch it. When GPT-6 writes a complex architecture with a flaw buried three layers deep, you won’t.

The better the model, the weaker the pilot.

The Way of the Shokunin

We don’t need to be anti-tech or anti-ai to solve this. We need to look at craftsmanship.

In Japan, a Shokunin (artisan) isn’t just a laborer. They are defined by a social obligation to do their best, regardless of whether anyone is watching. For the Digital Shokunin, the code isn’t just a product; it is a fingerprint of thought.

In the age of AI, “thought” doesn’t mean writing every loop. It means defining the Truth.

Manual review is a weak defence. The true Shokunin shifts craftsmanship from writing the implementation to building the deterministic guardrails, tests, and specs that the AI must fill.

The goal is to be the Architect of Truth.

The Middle Way: The “Cyborg Shokunin”

We are not going back to coding on stone tablets. We have deadlines. We have high-stakes projects that expect velocity.

The future belongs to the Cyborg Shokunin. You must use the exoskeleton to lift heavy weights, but never let your own muscles atrophy.

Practical Habits for the Augmented Age:

  1. AI-Free Fridays (Manual Mode): Athletes don’t wear their competition gear during strength training. Once a week, pick a complex ticket and turn off the AI. Write the logic naked. It will hurt, and it will be slow. That is what it means to practice.
  2. The Spec (Guardrails First) Manual review is brittle. To control the Agent, you must first define the Truth. Write the Spec and some Tests by hand. This forces you to crystallise the domain logic. Only then, let the Agent generate the implementation to pass your guardrails. If you cannot articulate the Spec, you do not understand the problem well enough to prompt it.
  3. Sparring, Not Outsourcing: Stop asking: “Solve this problem.” Start asking: “I’m planning to solve it using a factory pattern. Play the role of a hostile Senior Architect and tell me why that’s a bad idea.” Use the AI to challenge your assumptions, not to replace your thinking.

The Reality Check

We are at a crossroads.

One path leads to a future where we are merely supervisors of black boxes, terrified of an error message we didn’t generate and can’t understand.

The Digital Shokunin argues that someone still needs to know how the engine works, or we are all just passengers in a car driven by a statistical parrot.

The other path is more challenging. It requires you to slow down occasionally. It requires you to struggle with a problem that a machine could solve in seconds to prove to yourself that you still can.

In Cyberpunk: Edgerunners, Maine leaves David with one final piece of advice before the chrome consumes him. It is a warning disguised as encouragement:

“This is it. It’s the end of the line for me. But not for you. Fast is what you do best, ain’t it? Just keep running.” -- Maine

We have been given the ultimate speed. The AI is our Sandevistan. Use it. Run fast. But unlike Maine, make sure you never lose your grip on the wheel.

Choose the more challenging one. Don’t be the passenger in your own career. Use the “Chrome”. But keep the mind “Ganic”.

Further Reading