RAD from 1991 Is the Missing Workflow for AI Coding Agents

James Martin's Rapid Application Development (RAD) methodology, introduced in 1991, is surprisingly prescient for today's AI coding with agentic tools. It advocates lightweight planning, rapid prototyping, iterative construction, and verified cutover. The video argues that AI agents have made prototyping incredibly fast, but the missing piece is a precise spec.

Modern vibe coding accelerates the prototyping phase, but it introduces security risks. The example of an expense approval app shows that AI-generated code can miss critical rules like self-approval prevention. A study cited finds 45% of AI-generated code has security weaknesses. The prototype alone is insufficient for production.

The solution is spec-driven development: discoveries from the prototype are captured in a formal spec that includes business rules, security requirements, and acceptance criteria. That spec becomes the source of truth, and the generated code must pass tests derived from it. This mirrors RAD's original two halves: a precise description and code generation.

The implication is that programmers shift from writing code to writing specs and verifying results. Teams must adopt governance around AI-generated code, combining rapid prototyping with rigorous specification. The tradeoff is speed versus safety, but RAD's framework shows both can coexist with discipline.