unknown

AI coding assistants have created an unexpected productivity paradox: while they accelerate code generation, they demand significantly more upfront architectural planning and specification work. A 2025 study found developers using AI tools took 19% longer to complete tasks despite believing they were faster. The shift has driven adoption of spec-driven development, where teams create detailed planning artifacts (spec.md, tasks.md, CLAUDE.md files) before any code generation. Without proper specifications, teams encounter the "zig-zag problem" of iterative back-and-forth that wastes time. AI assistants lack contextual memory between sessions and suffer from context pollution during extended debugging, making clear architectural constraints essential. Code review times have increased 91% in high-AI-usage teams, and 45% of AI-generated code contains security vulnerabilities. Success requires treating AI as a junior developer that needs explicit guidance rather than a self-correcting autonomous agent.

The Architecture Tax: AI Coding Assistants Demand Better Planning

Exploring the intersections of artificial intelligence, decentralised cognition, posthuman ethics, society and culture.