Spec-Driven Development for LLM Coding: How to Avoid Chaos in Real Projects
Details
Large language models can generate code very quickly, but in real-world projects this often leads to unexpected changes, large pull requests, and difficult code reviews - especially in existing codebases with tests, multiple services, and legacy constraints.
The problem is usually not the model itself, but the lack of a clear workflow that controls what the agent is allowed to change.
In this talk I present a spec-driven approach to LLM-assisted development.
Instead of generating code directly, the developer first defines the change as a specification, validates the plan with the model, and only then allows the coding agent to implement it.
The specification acts as guardrails for the agent and also becomes a natural base for a clean pull request description.
We will look at a practical example from a backend / .NET style codebase, go through the full workflow, and discuss trade-offs, limitations, and when this approach is worth the extra effort.