Spezifikationsgetriebene Entwicklung
Definition
Spezifikationsgetriebene Entwicklung baut KI-Systeme (Agenten, Pipelines, Tools) aus expliziten Spezifikationen: requirements, output formats, allowed actions, and constraints. Specs are retrieved and used at runtime (z. B. in RDD) so behavior stays aligned with intent.
Es ist especially useful for agents and RDD: anstatt enProgrammierung all rules in weights or prompts, you maintain specs (z. B. in docs or a knowledge base) and retrieve them at runtime. Fits regulated domains and teams that want behavior to be auditable and updatable ohne Neutraining.
Funktionsweise
You write specs (natural language, schemas, or structured rules) and index them for Abruf (z. B. in a vector store or structured repo). At runtime, the task (und optional the current state) is used to retrieve relevant spec fragments. The model or agent decides (z. B. next step, allowed actions) and generates (output, tool call) mit dem spec in context. Validate checks the output against the spec (z. B. schema, rules); if validation fails, you can retry or surface an error. This keeps generation and Entscheidungs aligned mit dem spec without baking everything into prompt engineering or Feinabstimmung.
Anwendungsfälle
Spec-driven development passt, wenn behavior must stay aligned with retrievable requirements (RDD, compliance, or safety).
- Building agents that retrieve and follow specs (z. B. RDD pattern)
- Enforcing output format and constraints (JSON, allowed actions)
- Regulated or safety-critical flows where behavior must nachzuahmen requirements
Externe Dokumentation
- LangChain – Structured output — Enforcing output format from LLMs
- OpenAI – Structured outputs