Spec Driven AI Programming

Overview

Spec driven AI programming uses explicit requirements, acceptance criteria, and architecture constraints to guide AI assisted implementation. It differs from ad-hoc prompting because delivery is measured against defined outcomes.

Core Principles

• Start from a written specification
• Define acceptance tests before implementation
• Require traceability from requirement to code change
• Enforce review and validation gates

flowchart LR
    A[Specification] --> B[AI Draft]
    B --> C[Automated Tests]
    C --> D[Code Review]
    D --> E[Release]

Practical Example - Feature Contract First

A feature specification can define:

• API request and response schema
• Error codes and validation rules
• Performance targets

AI can generate initial code and test skeletons, but merge should remain blocked until contract tests pass.

public sealed record CreateInvoiceRequest(string CustomerId, decimal Amount);
public sealed record CreateInvoiceResponse(string InvoiceId, string Status);

public interface IInvoiceService
{
    Task<CreateInvoiceResponse> CreateAsync(CreateInvoiceRequest request, CancellationToken ct);
}

Production Example - Governance in CI

Teams using AI coding tools often add policy checks in CI:

• Pull request must include requirement reference
• Contract test suite must pass
• Security scan must report no new high severity issue
• Reviewer validates architecture boundaries

Common Pitfalls

• Prompting without a concrete specification
• Accepting generated code without deterministic checks
• No change log describing what AI generated

Key Takeaways

• The specification is the control plane for AI assisted implementation.
• Validation gates convert AI speed into safe delivery.
• Human review remains critical for architecture and risk.