Discussion: The role and value of Swagger Codegen in the age of AI / LLMs

[marclabud]

With the rapid rise of Large Language Models (LLMs) and AI coding assistants, a foundational question arises for modern architecture teams: Is deterministic specification-driven code generation still necessary, or can AI handle SDK generation entirely?

While AI is excellent at drafting code, swagger-codegen (and the wider OpenAPI ecosystem) remains a vital, highly valuable component of production-grade engineering. Instead of being replaced by AI, traditional code generators are becoming its ultimate pairing tool.

Here is a breakdown of why this paradigm remains critical:

1. Determinism vs. Probability (The "Contract" Layer)
   APIs are strict structural contracts. A type system cannot tolerate slight naming variations or missing fields.

The AI Risk: LLMs are inherently probabilistic. Even with strict system prompting, an AI generating a complex SDK from scratch can introduce subtle hallucinations, omit edge cases, or subtly alter payload mappings.

The Codegen Advantage: swagger-codegen is entirely deterministic. If the OpenAPI specification is valid, the generated client libraries and server stubs map perfectly every single time.

2. The Ideal Modern Workflow: "AI Authors, Codegen Compiles"
   The strongest architecture pattern today isn't choosing between AI and Codegen; it is combining them:

Design Phase (AI): Engineers leverage AI to reason through business logic, quickly draft, and optimize complex openapi.yaml or swagger.json files.

Execution Phase (Codegen): The AI-generated specification is fed into swagger-codegen within a CI/CD pipeline to flawlessly emit matching SDKs for TypeScript, Go, Java, or Python.

This blends the creative velocity of AI during the design phase with the bulletproof execution of traditional automation at deploy time.

3. Automation and CI/CD Efficiency
   AI pipelines are computationally expensive, suffer from variable latency, and are complex to integrate into rigid build automation loops. swagger-codegen runs instantly, locally or in a GitHub Action, costing virtually nothing to reliably regenerate a matrix of client libraries whenever an API definition changes.

4. AI Agents as Consumers of Specs
   Furthermore, AI agents themselves struggle to interact with arbitrary backends without hyper-rigid schemas. Tools like OpenAI Actions and the Model Context Protocol (MCP) rely on structured OpenAPI definitions to understand backend limits. We need Swagger to define the universe that AI agents operate within.

Conclusion
We shouldn't ask AI to build our SDK code from scratch; we should ask AI to help us write better API contracts, and rely on deterministic tools like swagger-codegen to build the infrastructure.

What are other teams' experiences? Are you injecting AI into your spec-to-code pipelines, or sticking entirely to traditional templates?