Kent Beck, creator of Test-Driven Development and signatory of the Agile Manifesto, publishes a thought-provoking LinkedIn post criticizing the "Vibe Coding" phenomenon: the practice of relying primarily on intuition and AI-generated code, without systematic testing discipline. His intervention, which carries considerable weight given his foundational contributions to software engineering, triggers a broad debate across the industry on the balance between AI assistance and engineering rigor.
Central argument: vibes are not enough
Beck's thesis: "if it feels right, ship it" is a fundamentally flawed approach, whether the code is human-written or AI-generated. Software correctness requires systematic validation: tests providing objective proof that the code works, regression prevention, executable specifications documenting expected behavior, and design feedback that surfaces complexity issues early. Intuition has value, but it does not replace rigorous verification.
TDD's relevance in the AI era
To those who consider TDD obsolete in the face of AI code generation, Beck responds: AI amplifies the consequences of undisciplined development. Coding faster without tests means accumulating technical debt faster; AI-generated code still contains bugs to catch; tests help refine prompts; TDD discipline prevents accepting plausible but incorrect AI output. TDD complements AI rather than being superseded by it.
A nuanced critique
Beck distinguishes: prototyping (vibe coding acceptable for throwaway experiments), production code (tests non-negotiable), personal projects (individual choice), and team or enterprise code (professional responsibility demands discipline). This is not a condemnation of AI assistance: the critique targets the abandonment of verification, not the use of tools.
AI as an amplifier
Key idea: AI amplifies existing tendencies. Disciplined engineers accelerate the red-green-refactor cycle with AI; undisciplined engineers produce buggy code faster. AI does not eliminate the need for discipline, it intensifies its consequences.
Reactions and common ground
Reactions split between traditionalists, pragmatists, and AI enthusiasts; counterarguments (TDD too slow, tests generatable by AI, market pressure) deserve consideration without invalidating the principle of verification. Consensus emerges around hybrid approaches: AI for code drafts and tedious test cases, tests validating AI output, rigor modulated by code criticality. Beck frames the issue in terms of professional ethics: AI changes how one codes, not the importance of quality.