Claude Code and nWave: Agentic AI That Craft

The Morning Everything Changed

I remember the exact morning. I was sitting in my home office, trying to implement a hexagonal architecture for a client’s payment system. The acceptance criteria were clear. The test was red. And I asked Claude Code to make it green.

What came back was not just working code. It was code that respected the port boundaries I had defined. It refused to mock inside the hexagon. It generated commit messages that followed my team’s conventions. For the first time, I felt like I was working with a colleague who had actually read the team’s engineering handbook.

That was the moment I understood that AI agents could enforce craftsmanship, not just produce code.

The nWave Orchestration

The nWave workflow emerged from a simple observation: AI agents are powerful but undisciplined. Left to their own devices, they will generate code that works but violates architectural constraints, skip tests that seem redundant, and produce commits that obscure rather than illuminate the development story.

nWave imposes structure. It defines five waves of work: Discuss, Design, Distill, Develop, and Deliver. Each wave has explicit inputs, outputs, and handover contracts. The orchestrator, Vera, ensures that no agent proceeds to implementation without approved acceptance criteria. No code ships without reviewer approval.

This is not bureaucracy. It is the same discipline that a senior craftsperson brings to a team, encoded into an agent workflow.

TDD at Machine Speed

The most powerful aspect of combining Claude Code with nWave is what happens during the Develop wave. The software crafter agent follows Outside In TDD with absolute fidelity. It writes a failing acceptance test. It writes failing unit tests through the driving port. It implements the minimum code to make them green. It commits.

Every cycle takes minutes instead of hours. But the discipline remains intact. The agent cannot skip the red phase. It cannot commit with failing tests. It cannot modify a test to make it pass. These constraints are not suggestions; they are hard coded into the agent’s protocol.

I have watched this agent produce, in a single afternoon, what would have taken me three days of careful work. Not because it types faster. Because it never loses focus, never gets distracted by Slack, never decides to refactor something shiny instead of finishing the current story.

Quality Gates That Cannot Be Bypassed

The nWave framework includes eleven quality gates that must pass before any commit proceeds. All tests green. No mocks inside the hexagon. Test count within budget. Business language verified. Static analysis clean.

A human developer might skip one of these checks when tired or under deadline pressure. The agent cannot. It runs every gate, every time, without exception. This consistency is perhaps the greatest contribution of AI to software craftsmanship: not creativity, but unwavering discipline.

The Human Still Matters

Before you conclude that developers are obsolete, consider what the agent cannot do. It cannot sit with a product owner and understand the unspoken fear behind a feature request. It cannot sense that a team is burning out. It cannot decide that the architecture needs to change because the business model has shifted.

The human craftsperson defines the vision. The AI agent executes with discipline. Together, they produce work that neither could achieve alone.

What We Learned

After six months of building with Claude Code and nWave agents, three lessons stand out. First, the quality of AI output is directly proportional to the quality of your acceptance criteria. Garbage in, garbage out applies to agents as much as it does to code. Second, agent workflows amplify your existing engineering culture. If your team values craftsmanship, the agents will enforce it. If your team cuts corners, the agents will cut faster. Third, the real productivity gain is not speed. It is consistency. Every commit follows the same standard. Every test verifies real behavior. Every review catches real issues.

The forge has a new apprentice. And this one never forgets the fundamentals.