Vibe Coding Still Needs a Craftsman

AI agents can write code faster than you ever will. That doesn't mean you can stop thinking.

AI agents can write code faster than you ever will. That doesn’t mean you can stop thinking.

My office has been deep in AI-assisted coding for a while now. Claude Code, agentic workflows, context-aware completions — the works. Productivity is up. Shipping speed is up. Everyone is excited.

And yet, some of the code that comes out of it is quietly terrifying.

It works. Mostly. But crack it open and you find a tangle of implicit dependencies, logic scattered across layers it has no business touching, and test coverage that’s either zero or cosmetic. It runs fine in the demo. Whether it survives six months of production changes is a different question.

After watching this pattern repeat a few times, I’ve started forming a clearer opinion: AI coding agents are a force multiplier. And force multipliers amplify what’s already there — the good and the bad.

Vibe Coding Has a Skill Floor

There’s a misconception that AI coding tools lower the barrier to entry to zero. That if you can describe what you want clearly enough, you’ll get working software.

You will get running software. Working is a higher bar.

No code experience means no ability to review what the AI produces. You can’t spot a hidden N+1 query, a missing transaction boundary, or a service that’s doing five things at once and calling itself “clean.” The code compiles, the tests pass if there are any, and it ships — until the edge case that wasn’t in the prompt surfaces at 2am.

Understanding code is still the job. AI doesn’t change that. It changes where you spend your time.

The Work Shifts, It Doesn’t Disappear

Here’s what actually happens when you add a capable AI agent to a development workflow: the cost of writing code drops dramatically. The bottleneck moves.

It moves to planning. To architecture. To understanding what you’re actually building before the first line is generated. Because the AI will happily build the wrong thing at incredible speed if you let it.

This is a good shift, actually. Writing code was never the hard part. The hard part was always figuring out what to write and how to structure it so you could change it later. AI just makes it more obvious that those are the things that matter.

Clean Architecture, TDD, DDD — these aren’t just process overhead that slows you down. They’re load-bearing walls. Remove them and the structure falls faster, not slower, because now you’re generating technical debt at machine speed.

DDD Makes the AI a Better Collaborator

This one surprised me when I first noticed it.

Domain-Driven Design forces your code to speak the language of the problem domain. Entities, value objects, aggregates, bounded contexts — these are names and boundaries that mirror how the business actually works. The code reads like documentation.

When you feed that kind of codebase to an AI agent, something useful happens: the AI can actually reason about it. It understands that an Order can’t be mutated after it’s been Confirmed. It knows that a Payment belongs to a bounded context that doesn’t reach into Inventory. It has vocabulary to work with.

Compare that to a codebase where a UserManager handles authentication, profile updates, notification preferences, and billing status. The AI has no conceptual map. It guesses. Sometimes well, usually not.

DDD doesn’t just help humans read code. It helps AI write better code.

TDD Keeps the AI Honest

AI-generated code has a particular failure mode: it looks correct. It’s syntactically clean, it handles the happy path, and it’s confident. The tests, if they exist, often test the implementation rather than the behavior.

Test-Driven Development flips the order. You define the expected behavior first — in executable form. Then the implementation exists to satisfy those tests. The AI can write the implementation, but the tests are the spec, and the tests were written by a human who understood what “done” actually means.

There’s a second benefit: the test suite becomes living documentation. Not a README that drifts out of sync over time, not inline comments that lie about what the code does — actual runnable proof of what the system is supposed to do. Six months from now, when nobody remembers why a particular rule exists, the test case will still be there explaining it.

Give the Agent a Conscience

The final piece is tooling — and there’s a specific one worth knowing about: Superpowers by Jesse Vincent.

Superpowers is an agentic skills framework — a collection of composable skill files that you install into your coding agent once, and from that point forward, the agent follows them automatically. No prompting required. The skills trigger on their own based on what you’re doing.

The workflow it enforces is exactly what good software development looks like:

  1. Brainstorming first — before touching code, the agent asks what you’re actually trying to build, teases out a spec, and shows it to you in digestible chunks for validation.
  2. Writing a plan — once the design is approved, it produces a detailed implementation plan broken into 2-5 minute tasks, with exact file paths and verification steps. Clear enough for a junior engineer to follow.
  3. TDD throughout — during implementation, it enforces red/green/refactor. Write a failing test. Watch it fail. Write the minimal code. Watch it pass. Commit. Code written before tests gets deleted.
  4. Subagent-driven execution — each task gets a fresh subagent with a two-stage review: spec compliance first, then code quality. It’s not uncommon for Claude to run autonomously for hours without going off-plan.
  5. Code review between tasks — issues are reported by severity. Critical issues block progress.

It works with Claude Code, Cursor, OpenCode, Codex, and Gemini CLI. Installation is a single command.

The point isn’t the tool itself — the point is the principle behind it. An agent without structure will take the path of least resistance every time. An agent with well-defined skills follows the process you’d want a human developer to follow: think before coding, test before shipping, review before merging.

That’s the difference between AI that helps you build something solid and AI that helps you dig a hole faster.

The Craftsman Isn’t Replaced

The pitch for AI coding tools is usually about speed. More features, faster shipping, less time writing boilerplate. That’s all real.

But the thing that doesn’t get said enough: the value of a developer who understands what they’re building, can reason about tradeoffs, and knows how to structure a system for change — that value just went up. Because now that developer can execute at a pace that used to require a team.

The craftsman isn’t replaced. The craftsman gets leverage.

Use it carefully.