𝗧𝗵𝗲 𝗖𝗼𝗱𝗲 𝗜𝘀 𝗖𝗵𝗲𝗮𝗽. 𝗧𝗵𝗲 𝗦𝗽𝗲𝗰 𝗜𝘀 𝗧𝗵𝗲 𝗔𝘀𝘀𝗲𝘁.
Code is becoming a cheap artifact. The real value now lives in the specification.
I spend less time writing implementation plans by hand. I spend more time on design. AI makes this possible. It does not replace engineering judgment. It changes where you use it.
I let AI draft specifications and code. My job is now defining intent and identifying constraints. The writing is the least valuable part of the process.
My specifications are not for humans reading a wiki. They are for the next AI session. They must allow an AI to continue work without new explanations.
Effective specifications focus on:
- Requirements
- Constraints
- Acceptance criteria
- Verification steps
They are built to be executed, not just read. The audience is the next contributor, whether human or AI agent.
Modern engineering is a constraint management problem. AI works well with constraints if you capture them clearly. My workflow follows these gates: Intent → AI Specification → Human Review → AI Implementation Plan → Human Review → AI Code Generation → Testing
I provide the goal, requirements, and boundaries. AI drafts the spec. I review it. AI drafts the plan. I review it. Only then do we generate code.
I write less, but I review more carefully. This is where the engineering value stays.
A good specification defines what must be true, not how to make it true. For example, a refactoring spec should state:
- No class in the application layer may reference DAO implementations.
- Acceptance criteria: Layering violations return zero matches during searches.
The most important task is identifying load-bearing constraints. These are critical rules like:
- Database initialization strategies
- Deployment models
- Integration boundaries
If you miss these, the system breaks.
AI sessions are temporary. They come and go. The value comes from the shared memory:
- Specifications
- Implementation plans
- Architecture Decision Records (ADRs)
- Conventions
This memory prevents documentation drift. When your README, code, and ADRs tell different stories, trust breaks. You must align them with reality.
Das Repository sollte diese Struktur widerspiegeln:
- CLAUDE.md: Der Workflow und die Review-Gates.
- status.md: Ein lebendiger Index aller Spezifikationen und Pläne.
- specs/: Das „Was“ und „Warum“.
- plan/: Das „Wie“.
- rules/: Coding-Konventionen auf Klassenebene.
- docs/adr/: Unveränderliche Aufzeichnungen wichtiger Entscheidungen.
KI kann Code generieren. Sie kann jedoch nicht zuverlässig bestimmen, welche Einschränkungen für Ihr Unternehmen relevant sind. Das liegt in Ihrer Verantwortung.
Bauen Sie ausführbares Wissen auf. Beginnen Sie jedes Projekt mit einem gemeinsamen Gedächtnis, nicht mit einem leeren Blatt.
Quelle: https://dev.to/daniel_wu_cac679a2760ba0a/the-code-is-cheap-artifact-now-the-spec-is-the-asset-3b02
Optionale Lern-Community: https://t.me/GyaanSetuAi