Vibe coding: the spec-to-code boundary dissolves
"Vibe coding" — a term popularised by Andrej Karpathy in early 2025 — describes a new way of developing: describe what you want in natural language, let AI generate the code, adjust through successive conversational iterations, without necessarily understanding the generated code line by line.
For Business Analysts, this evolution raises a fascinating question: if functional specifications can directly generate working code, what happens to our role?
Direct spec → code connection: reality or illusion?
The promise: write "I want a leave request form that checks balances, sends a manager notification and updates the HR calendar" and get a working application in minutes.
The 2025 reality: it's partially true. For discrete, well-defined features without integration complexity, tools like Cursor, Claude Code or Replit Agent can generate working code from a natural language description. But as soon as you touch existing systems, architecture constraints or subtle business rules, quality drops rapidly.
What vibe coding changes for BAs
Specification precision becomes critical
Paradoxically, vibe coding increases the importance of quality specifications. AI generates exactly what you describe — if the description is ambiguous, the code will be functionally incorrect. Edge cases, performance constraints, implicit business rules must now be made explicit in specifications, because AI won't guess them.
The BA becomes functional quality validator
If a developer can prototype a feature in 2 hours through vibe coding, the BA must be able to validate it functionally in 30 minutes. This requires understanding how to test nominal cases and edge cases, and mastering testing tools.
The new role: "Prompt Architect"
"The 2026 Business Analyst won't write specs for human developers anymore. They'll write structured prompts for AI development agents, while maintaining responsibility for functional quality."
Specification-writing skills don't disappear — they shift towards writing rich system prompts, structured development contexts, and formalised validation criteria.
