unknown

AI accelerates execution so dramatically that the slow, deliberate phases of software development — requirements clarification, design, problem definition — become more valuable, not less. Using Kahneman's System 1/System 2 framework, the argument is that LLMs are essentially fast pattern-matching (System 1), while deciding what to build and why still requires human judgment (System 2). When execution is cheap, mistakes in upstream decisions propagate faster and cost more. Practical techniques include writing down the problem before prompting AI, running AI-assisted pre-mortems, building throwaway prototypes to validate direction, and surfacing edge cases before implementation. Managing stakeholder pressure around AI-driven velocity expectations is also addressed, with advice on making the 'slow phase' visible through artifacts and timeboxing.